Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
113 prevalence of carriaqe of various pathoqens by farm animals : ( b ) the antimicrobial susceptibi l ity patterns of these pathoqens : ( c ) the prevalence of infections caused by these pathoqens in humans . Data do not exist to answer directly the princ ipal question posed to the committee even for a wel l -recoqni zed pathoqen such as Salmonella spp . Indeed , the data are sparse and confl ictinq even with reqard to the subordinate quest ions cited above . To i l lustrate some of the problems confront inq the committee , Fiqures VI I - 1 and VI I - lA provide a summa ry of current information about the impact of druq resistance of salmonel lae on the epidemiol oqy of salmonellosis in human or animal populations exposed or not exposed to antimicrobial aqents . In constructinq Fiqures VI I - 1 and VI I - lA , it was assumed that there is a qradual shi ft of stra ins from druq susceptib il ity to druq res i stance . The committee bel ieves that druq res istance is a mani festation primarily of exposure of bacteria to antimicrobial aqents for lonq periods with ultimate selection of res istant stra ins : thus , it is antic ipated that with increas inq time of exposure , the prevalence of res i stant stra ins in the animal and human populations wi l l increase . For certa in elements in both Fiqures VI I - 1 and VI l - lA , no data are ava i lable , as indicated in parentheses a fter the item . The studies cited in Fiqures VI I - 1 and VI I - lA are l imited in appl icabil ity by the fact that they were not done as part of a cohes ive attempt to address the overall issues posed to this committee , but rather were done to address more l imited aspects of the problem . As i l lustrated in the first hori zontal l ine of Fiqure VI I - 1 , the "maj ority" of the reports show that salmone l l ae in the fecal flora of farm anima l s are res i stant ( i . e . , res i stant to at least one antimicrobial ) and a minority are susceptible (meaninq susceptible to ampicill in or the tetracycl ines ) . By contrast , the maj ority of the reports show that human isolates are sti l l susceptible to commonly tested antimicrobials ( Fiqure VI I - lA) . The prevalence of res i stance appears to be ris inq both for ï¿½ Â£Qli and Salmone l la spp . No data prove directly that administerinq antimicrob ial aqents in subtherapeutic doses to farm anima l s increases the prevalence of carriaqe of susceptible salmonel l a in farm anima l s : the arqument (a hiqhly unl ikely one ) would be indirect by analoqy with the ef fect of antimicrobial aqents on infect ions by enterohemorrhaqic ï¿½ col i ( EHEC ) in humans . ? By contrast , the 19 8 0 NRC report 2 1 cited various studies showinq that the feedinq of antimicrob ial druqs to farm animals enhanced the rate of el imination of susceptible stra ins of Salmonella spp . : thi s effect would result in a decrease in the prevalence of these susceptible isolates .
Susceptible Strains Resistant Strains 1. Current prevalence Minority of strains MaJority of strains ++ (numerous studies of E. coll in animals [see text] but for Salmonella, 2. Effect of subtherapeutic + (no data except by analogy and extra potation from administration on prevalence with EHEC** strains In humans, "etiologic fraction" concept see text) In humans) - (1 980 NAS 21 ) - (no data) o (Fagerberg 13 ) FIGURE VI I - 1 . Potential Ef fects o f Ant imicrobial Use on Prevalence o f AntimicrobialÂ ï¿½ Susceptible and -Resistant Sa lmone l la Stra ins in Farm Animals . Figure prepared by the committee . + represents mild increase in preva l ence of stra ins , in degree of virulence , or in the characteristic spec i f ied : ++ represents a moderate increase : and +++ represents a maj or increase . - represents a mild decrease in preva lence of stra ins , in degree of virulence or in the character spec i f ied : - - represents a moderate decrease : and --- represents a maj or decrease . o represents no change . * * EHEC = Enterohemorrhagic ï¿½ QQli .
Susceptible Strains Resistant Strains 1. Cu rrent prevalence Majority of strains Minority of strains 2. VIrulence for humans Current level + compared to susceptible strains (abUity to colonize and cause (see Chap Ill) disease) + (by analogy with EHEC)** +++ \etiologic fractionâ¢; see text) 3. VIrulence for humans -- (by anal ogy with effect on - (no data) taking antibiotics clearance In animals per for other reasons 1 980 NAS report 21 (great d iffic ulty If "wrongâ¢ drug chosen, presumably uncommon) 4. Efflency of treatment Current effectiveness of Infections ..... **EHEC = Enterohemonhaglc E. coli. ï¿½ FIGURE VI I - la . Potential E ffects of Antimicrob ial Characteristics on AntimicrobialÂ susceptibl e and -Resistant Sa lmonel la Stra ins in Humans . Figure prepared by the committee . + represents mild increase in preva lence o f stra ins , in degree of virulence , or in the characteristic spec i f ied ; ++ represents a moderate increase ; and +++ represents a maj or increase . - represents a mild decrease in preval ence of stra ins , in degree of virulence or in the character spec i f ied ; - - represents a moderate decrease ; and --- represents a maj or decrease . o represents no change .
116 There are few data o n animals concerning the effect o f feeding subtherapeutic doses of antimicrobial agents o n the carriage rate of drug-res istant stra ins . However , Bohnhoff and col leagues showed many years ago that the feeding o f a single oral dose of streptomycin to mice markedly increased their susceptibil ity to infection by a stregtomyc in-resi stant stra in of salmonel lae administered orally . 5 S imilar findings were later reported by Meyne11 , 2 0a Bohnhoff and Mi l l er , 5b Meynel l and Sabbaiah , 2 0b and Mil ler and Bohnhoff . 2 0c Furthermore , by extrapolation from the " etiologic fract ion" in humans ( the proport ion of infections that would not have occurred but for the res istance of the infecting bacterial strain to the ant imicrobial administered ) , one would expect a marked enhancement of infectivity and hence of prevalence . No data support a diminution in the prevalence of drug-res i stant stra ins as a result o f feeding subtherapeutic doses o f antimicrobial agents . One study by Fagerberg 1 3 indicates no d i f ference in the cl earance rates of tetracycl ine-resistant stra ins o f sa lmonel l ae between anima l s given tetracycl ine and those given another ant ibacterial drug . I n assess ing the second , third , and fourth elements of Figure VI I - 1A , which deal with the impact o f salmonella infections in humans , the committee had the opinion ( see below) that the maj ority of stra ins of salmonel lae that find thei r way into humans are transmitted from food products which originate on the farm . The second element dea l s with the ef fect o f antimicrobial res istance on the virulence o f sa lmone l l ae for humans . Various authors have used the term "virulence " in dif ferent ways . Some have restricted the term to the abi l ity to cause disease , particularly toxin-med iated d i sease , whereas others have incorporated the abil ity to col on i z e and to cause disease by any mechani sm . The committee decided to use the second defin ition for thi s assessment and t o use the terms "virulence" and " infectivity" interchangeably . I f drug-susceptible stra ins are cons idered as a basel ine , there is evidence ( see Chapter I I I ) that drug res istance may be associated with either a decrease or an increase in virulence . On balance , the committee dec ided that the data were more compel l ing for either no change in virul ence or an increase in virulence than they were for a decrease in virul ence , although the data are weak and rather inconclus ive . In terms of the virulence of Salmonella spp . for humans who are taking antibiotics for other reasons ( el ement 3 in Figure VI I - 1A ) , there is strong evidence that drug res i stance of the salmonellae faci l itates infection . In persons in this category , whose disease i s included in the " etiologic fraction , " drug-resistant stra ins are able to colon i z e the gastrointestinal tract and cause disease even in inocula too sma l l to cause infection in other circumstances , presumably
117 because the antimicrobial drugs inh ibit the normal competing flora ( see below ) . By contrast , the committee is aware o f no data indicating that drug resistance diminishes the . infectivity of Salmonella spp . for humans . The committee also is unaware of any data on the effect of taking ant imicrobial agents on the infectivity of drug-suscept ible stra ins . However , by ana l ogy with data in anima l s noted under element 2 in Figure VI I - 1 , the admini stration of ant imicrobial agents could enhance the rate of el imination and hence reduce the infectivity o f drug-susceptible stra ins for humans . Nevertheless , us ing infection by enterohemorrhaï¿½ ic ï¿½ Q21i ( EHEC ) as an analogy , a study in 19 8 7 by Carter of EHEC infection in a nurs ing home showed a higher rate o f secondary infection among patients who were taking antimicrobial agents to which EHEC was presumably susceptible ( isolates of EHEC appear to be almost uni formly susceptible to ampicil l in , tetracycl ines , chl oramphenicol , and trimethoprim-sul famethoxazole , 2 2 a , 2 3 , 2 9 a as opposed to a control group that was not taking such antimicrob ial agents . Thi s suggests a facil itating effect of antimicrobial agents even upon infection by susceptible stra ins . F inal ly , the committee cons idered the ef fect o f drug res istance on the treatab i l ity of salmonella infections in humans . In princ iple , res istance should lead to increased d i f ficulty in treatment . However , cons idering the epidemiology and the population at greatest risk of death i . e . , neonates and the very elderly--it seemed l ikely to the committee that many patients who die of salmone l l o s i s never receive speci fic antimicrob ial treatment and that fa ilure o f treatment because the wrong drug was chosen may b e uncommon . However , some patients may be treated inadvertently because the i r phys ician does not recognize the infection as salmonel losis . Treatment that prevents bacteremia , a rare event anyway , might lead to an unrecogni zed bene fit . Taking these considerations into account , the committee cons idered that drug res i stance is an uncommon cause of treatment fai lure . overall , the committee concluded that the maj or consequences of feed ing antimicrob ial agents to anima l s or humans are l ikely to be : 1 ) a tendency to increase the prevalence of drug-resi stant strains : 2 ) an e f fect on both the pathogen and the fecal f lora that might alter their usual interaction : and , thus , the rel ative infectivity o f the pathogen . The number o f reported cases of salmonel losis in the u . s . has r isen progressively over the past three decades , a period during which the practice of subtherapeutic administration o f antimicrobial agents to farm animals has been stead i ly increas ing ( Figure VII -2 ) . However , thi s observation does not prove that the increase in salmonellosis is rel ated to antibiotic use because other potentially
z 0 t= j ï¿½ 12 0 0. 0 10 8 / -"' ..... 0 â¢ \ ..... e - , , , ClO 0 - / , , / ï¿½ ... 6 ' / ' ï¿½ .., ï¿½ â¢ 2 0 ' 1 S aQ 1 S&S 1 170 1 1 75 1 S80 1 gss ted to the Cente rs for FIGURE VI I-2 . Salmonel l a ( - ) and Shig e l l a ( -- ) infec tion s repor 0 , 0 0 0 popu latio n in the Unit ed State s . Disea se Control , 1 9 5 5 - 1 9 8 4 . Rates are per 1 0 Repr inted from Cha lker and lla typh i . Sa lmone l la rate exclu des infec tion s due to Sa lmone Bla ser . 1 1 ,I
119 confoundinq variables have occurred in the same interval , i ncludinq the increas inq use of convenience foods and prepared " fast foods . " That the increase in reported cases of salmonellos is in humans over the past 3 0 years is not an art i fact of reportinq is suqqested by the fact that the number of infections caused by Shigella spp . has rema ined fai rly constant durinq thi s period ( Fiqure VI I -2 ) . Thus , there i s direct evidence of an increase in salmonellosis in humans . Some of thi s increase miqht be attributed to some of the elements shown in Fiqure VI I - 1A . Furthermore , it i s di f ficult t o deteï¿½ine directly the contribution o f druq res istance to fatal ity from salmonella infection because ne ither the pub l i shed CDC data for reported cases nor the NCHS f iqures for deaths from salmonellos i s make note of the druq suscept ibi l ity or res istance of the pathoqen . After cons ideration of the concepts il lustrated in Fiqure VI I - 1 and the l imited data avai l able perta ininq to the issues , the committee concluded that it was impossible to arrive at a firm answer to the important quest ion of whether or not the administration of ant imicrobial aqents in subtherapeutic doses to farm animals has led to an overal l chanqe in the total number of cases of sa lmonellos i s i n humans . Accordinqly , the committee decided to approach the probl em indirectly by devis inq a risk model that focuses upon one aspect of the problem , namely , the number of deaths which can be attributed to the subtherapeutic administrat ion of ant ibiotics to farm anima l s . STRUCTURE AND LIMITATIONS OF THE MODEL As summari zed above , questions about the s i ze of the human risks from low-level farm uses of ant ibiotics cannot be answered by the direct interpretation of data on thi s matter . There fore , the committee developed and adopted a conceptual approach , or model , in which some informat ion is ava i l able at each step . In devis inq the model , the committee chose to dea l only with salmonel los i s because this was the only pathoqen for which there were data avai l able in quant ity and qual ity that the committee could use in quanti fyinq the risk . Nevertheless , the committee recoqnized that there are other infect ious orqanisms that may account for at least as larqe a part of the overal l problem of human i l l ness attributable to the subtherapeut ic use of antibiotics on the farm . The model includes a sequence o f f ive quant itative estimates , each dependent on the prior estimates . In steps 2 , 3 , and 5 ( see below ) , the estimates were calculated separately for res istance to any antibiotic and for res istance to at least penici l l in/ ampici l l in or the tetracycl ines . These estimates are i l lustrated in parentheses by the committee ' s mid-ranqe estimate for each
12 0 step with respect to resistance to penici l l in/ ampici l l in or the tetracycl ines , as in the fol lowing : 1 . Annual number o f cases o f salmonel los is reported i n the u.s. ( 50 , 000) . 2 . Fraction of human cases due to bacterial stra ins showing res istance to penici l l in/ampicill in or the tetracycl ines ( 15 % ) . 3 . Death rate ( 1 . 0 % ) among cases with drug res istant - salmonellosis . 4 . Fraction o f these deaths associated with infection by bacterial strains o f farm origin ( 7 0 % ) . 5 . Proportion o f this fraction resulting from subtherapeutic use of penici l l in/ampici l l in or the tetracycl ines in animal feed ( 9 0% ) . S ince these estimates are l inked in stepwise fashion , and each is developed to be statistical ly conditional on a l l that precede it , they can b e multipl ied to estimate the number of deaths . Thi s chapter develops the risk model and explains the committee ' s choices of the quantitative inputs the model requires , and Chapter VI I I uses the model to develop estimates of excess deaths while reflecting the uncertainty o f those estimates . FOcuS ON SALMQNELLAE The model has several l imitations . one l imitation i s that i t assesses only the ha zards of infection with Salmonella spp . The maj or foodborne pathogens known or suspected to be transmis s ible to humans from farm anima l s o r their products are Salmonel la spp . ; Campyl obacter j ej un i enterohemorrhagic stra ins o f Escheri ch i a col i ( EHEC ) , especi a l ly serotype 0 1 5 7 : H7 ; and Yers i n i a enteroco l it i c a . The Centers for Disease Control reported on 1 5 1 outb re a ks o f foodborne bacterial disease i n 1 9 8 2 : 8 , 9 of this number , salmonellae caused 5 5 outbreaks with 2 , 0 5 6 cases and 8 deaths . By contrast , each of the other three species caused only two outbreaks , with between 3 1 and 1 8 8 cases for each species , and no deaths . However , these data are highly sel ected and are incomplete . Although recent surveys 3 , 4 , 5 , 14 , 15 , 19 , 2 6 , 2 7 , 2 8 suggest that Campyl9bacter spp . may cause at least as much i l lness a s does salmonellos i s , data o n infections caused b y nine named or proposed named campylobacter species are not ava i l able . Laboratory-based national surve i l l ance of campylobacter
12 1 infections in the u . s . began in 1 9 8 2 with a panel o f 1 1 states with additions in 1 9 8 3 bringing the total t o 3 1 states . 1 0 The committee has not critica l ly looked a t these data . Also , antimicrobial res istance i s not recoqn i z ed as a maj or issue in treatment , and no data are avai l able to determine whether the administration of ant ib iotics to farm animals has contributed to antimicrobial res istance in these spec ies . S imilarly , there are no nationwide data ava i l able to estimate the impact of infection caused by x . enterocol it i ca or EHEC . However , infection caused by ï¿½ . enteroco l it ic a i s rarely recoqnized i n the United States . Although in fect i ons may be grossly underdiagnosed , more data wou ld be n e ed e d t o estab l i sh yers iniosi s as an important cl inical problem in thi s country . Most isol ates of ï¿½. enterocolitica are susceptible to tetracycl ine , but ampici l l in res istance i s common . 6 Stra ins of EHEC are almost always susceptible to commonly used antimicrobial agents . 2 2 a , 2 3 , 2 9 a There are over 1 5 0 0 serotypes of salmonel lae . 12 However , more than 7 0% of the infections are caused by 10 serotypes and four are dominant : ï¿½. typhimurium causes about one-third . of reported infections , ï¿½ . enteritidis about 1 0 % , ï¿½ . heidelberg about 1 0 % and ï¿½ newport about 5 % . The frequency of i solation of ï¿½ . typhi , which causes typho id fever , has diminished sharply s ince the beginning of thi s century ; currently , there are about 5 0 0 isolat ions o f ï¿½ . typhi per year , as opposed to more than 4 0 , 0 0 0 isolations of other species of salmonellae . ï¿½ . typhi i s not known to infect anima l s or to have an animal reservoir and is thought to be spread from person to person , so it i s not considered in thi s analys i s . The committee concluded that Salmonella i s the only genus for which sufficient data are avai l able to est imate the national impact on mortal ity from infect ions caused by drug-resistant organisms transmitted from farm anima l s or their products to humans . The rema inder of the quantitative analys i s in thi s report pertains to infection caused by nontypho idal Salmone l l a spp . Morbidity was not included in the risk calculations . MODEL QNCERTAINTY A second l imitat ion is that the model itsel f may be incorrect . While the steps outlined above are logical ly appeal ing , other chains of critical events could be developed , such as steps in the cha in of transmiss ion or pathogenes i s , and these might produce material ly d i fferent est imates . Numerous d i f ficulties , conceptual and practical , impede the estimation of the mortal ity rates attributable to salmonella infections , and the committee recommends a
122 substantial increase in the investiqation and development o f conceptual model s of thi s matter that include morbidity . INDEPENDENCE OF ESTIMATES Third , our use of the model requi res that the parameter estimates be conditionally independent ; that i s , the statistical distribution of one estimate , qiven others that precede it in the model , depends on those values in a way that is ful ly speci fied by the model . For example , the death rate amonq cases with druq-resistant salmonellosis ( see step 3 in Table VI I I - 1 ) re fers to reported cases in the u . s . ; it is presumed that there is a much larqer number of unreported cases ( see below ) . The death rate is l ikely to be lower for the unreported cases than for reported cases . However , it i s not qenera l ly poss ible to val idate thi s assumption o f conditioned independence from available data for salmonel l a risks . Further , l imitations in the data have required the use of some estimates that are not conditional , or that are less completely condit ional , than the model theoretica l ly requires . Both the numerators and the denominators needed to calcul ate the rates of il lness and death are subj ect to cons iderable uncerta inty . Further , these uncerta int ies are closely l inked , so that the numerator and denominator must be developed toqether and the result inq ( death ) rate should be appl ied to other settinqs only inso far as the maj or uncerta inties are appropriately correlated . For example , the death rate of salmone l l a infections depends very much on whether it is for cases such as are reported to the Centers for Disease Control or for the whol e of symptomatic infections i n humans . Thus , the seven deaths reported amonq 5 0 3 pa tie ï¿½ts in a 1 9 7 9 - 1 9 8 0 CDC survey yield a death rate of 1 . 4 % . 1 2 S 1nce the denom i nator was 5 0 3 cases reported in approximately rout ine fashion to the CDC , one miqht , with caut ion , estimate that about 1 . 4 % of al l such reported cases of sa lmonellosis miqht have been fatal . Because about 5 0 , 0 0 0 cases of sa lmonellosis are reported to the CDC annual ly , one could make an estimate that there are about 7 0 0 deaths in 5 0 , 0 0 0 reported cases per year . In fact , it is reasonable to assume there are many more unreported cases ; however , the death rate estimate of 1 . 4 % does not necessarily need to be chanqed , because , unreported cases miqht be less l ikely to be severe or fatal and to qo unreported for that reason . S imilar cons iderations apply to other aspects of salmonella infect ion , such as the rate of hospita l i z at ion , the proport ion of patients with " serious " rather than mild disease , and a med ical decis ion to culture stools or other materials .
123 LIMITATION TO CQC-BEPQRTED CASES OF SALMQNELLQS I S The committee recogni zes that the numbers of cases of salmonellosis reported to CDC per year , ranging from about 4 0 , 0 0 0 to 6 5 , 0 0 0 , surely is an underestimate of the number of cases in the u . s . ; nevertheless , the committee decided to use thi s range o f numbers for the first step in the risk assessment . Thi s deci s ion was made because : 1 ) several other critical estimates in the risk model apply to thi s same populat ion of CDC-reported case , and 2 ) it may be that unreported cases are mi lder and of lesser consequence , although thi s has not been shown to be the case . However , i nvestigation o f epidemics by CDC o f reported cases necessarily underestimates the scope of the probl em . LIMITATION TO ESTIMAtES OF MORTALITY A fi fth l imitation is that the model dea l s only with lethal infections . Salmonel lae also cause cons iderable , albeit temporary , personal d istress (morb idity ) , as wel l as a large economic burden . The cl inical mani festations range from asymptomatic colon i z ation through mild or somet imes severe diarrhea , to disseminated and somet imes l ethal il lnesses , such as meningitis or osteomyelitis . However , statistical data regarding the incidence o f various symptoms are minimal or lacking even for severe cases , and there is d i f ficulty in applying such data to the u . s . as a whole . Fatal ities due to salmonella infections are clustered in the very young and in the elderly ; Table VI I - 1 shows the age distribution of salmonella deaths reported to the National Center for Health Statist ics ( NCHS ) for the years 1 9 6 8 - 19 8 5 . For each end point , the group at risk must be unambiguously def i ned , essentially all instances in the group must be ident i fied , and the calculated rate must be appl ied to other groups only when they are l ikely to have about the same distribution of severity of i l lness and only when adequate margins of error are attached to the ca lculated rate . These margins of error wi l l ordinarily be substant ial ly wider than stati stical conf idence l imits . The hospita l i z at ion rate for salmonellos is in the 1 9 7 9 - 19 8 0 survey of selected communities was 4 5 % and the rate in the 1 9 8 4 - 1 9 8 5 survey was reportedly s imilar . 12 However , in a revi ew of recent outbreaks of ï¿½ . enterit id i s infect ions oc c urring in the northeastern u . s . , the hospital i z at ion rate was estimated to be only 12 t . 3 0 Thus , a range of 1 2 % to 4 5 % may b e entertained a s the estimate for the rate o f hospital i z ations of patients with salmonellos i s . When these rates are appl ied to the 5 0 , 0 0 0 reported cases of salmonellosis per year , the number of pat ients hosp ital i zed for thi s infection ranges from 6 , 0 0 0 to 2 2 , 5 0 0 per year .
124 TABLE VII - 1 FREQUENCY AN D PERCENTAGE O F DEATHS DUE TO SALMONELLOS IS ( By Age , for 1 9 6 8 - 19 8 5 ) ï¿½Atbl Pe rce nt Age Number Per YeAr of Age Under 1 day 1 0.1 1 - 6 days 8 0.6 7 - 2 7 days 30 2.1 2 8 - 3 64 days 165 11 . 6 1 - 4 years 42 3.0 5 - 9 years 12 0.8 1 0 - 14 years 11 0.8 1 5 - 2 4 years 14 1.0 2 5 - 3 4 years 30 2.1 3 5 - 4 4 years 42 3.0 4 5 - 5 4 years 104 7.3 5 5 - 64 years 176 12 . 4 6 5 - 7 4 years 3 14 22 . 1 7 5 - 8 4 years 296 20 . 8 8 5 + years 174 12 . 2 Unknown 2 0.1 Al l Ages 1 , 421 100 . 0 Source : National Center for Health Stat i stics . 22
125 LIMITATIONS IN THE Dill S ixth , the model i s l imited by the ranqe and qual ity of data avai lable for the estimates required , as summari zed above . The committee has developed three estimates for each quantitative parameter : a mid-range estimate , a high estimate and a low estimate . The mid-range estimate expresses the committee ' s best j udgment about the value that is equa l ly l ikely to be too large or too smal l --a median , o f sorts , of the committee ' s col lective obj ective and subj ective j udgment . The high and low estimates express the committee ' s best j udgments about the range of fiqures that most other experts would find plaus ible . These l imits are not presented as stati stical confidence l imits ( even subj ect ively ) , nor as outs ide bounds on poss ibil ity . For example , if three f iqures for some parameter were 1 and 1 0 , an estimate outs ide the range 1 , 3 , and 1 0 would not be credible , the committee bel ieves , to most other experts . We have not attempted to attach probabi l ity values to these low and high est imates , because we have no direct evidence about what l imits other experts would be will ing to accept as plaus ible . The committee d iscussed at some length how these three est imates should be derived . The bas ic point of discuss ion was the extent to which our col lective subj ective j udgment should be used to modi fy speci fic values obtained from the l iterature . As an example , the fraction of stra ins res istant to two or more antib iot ics has been rising , in contrast to publ ished results that necessarily refer to infections detected in the past . Thus , the publ ished range of rates of res istance to multiple antibiotics w i l l tend to be too low , but by an unknown amount . How much should the committee ' s j udgment about this trend over time be integrated into rates obtained from the l iterature? Other problems arise because of the need to use est imates that are statistical ly conditioned on preceding estimates , though appropriate data may not exist . For example , the fraction of infect ions due to multiresistant stra ins should refer spec i fica l ly to the kinds of cases , with speci fic deta i l s , reported to CDC : however , not all sources meet thi s requ irement . In the end , the committee tended to give its subj ective j udgment cons iderable weight . We have not attempted to attach probabi l ity values to these low and h igh estimates , because we have no direct evidence about what l imits other experts would be wi l l ing to accept as plaus ible . The remainder of th is chapter outl ines the bas i s for the est imates that the committee used in its risk model to assess the contribution of subtherapeut ic use of antibiotics in animal feed to the presence of drug-res istant salmone l l ae in humans in the United States .
126 NUMBER OF CA5ES PER YEAR The number of cases of salmonella infections per year is large , but not precisely determined . Many observers bel ieve that it is probably 1 0 - 1 0 0 times larger than the number of confirmed cases reported to the CDC . Thi s i s because many patients with salmonellosis do not seek medical attent ion : when they do , stool s or other specimens may not be cultured : when cultures are attempted , they may be unsuccessful in isolating the infecting bacteria , or pos itive results may not be reported to CDC . Stil l , most of the data relating to morbidity and mortal ity from salmonel losi s in the u . s . are derived from the CDC . The CDC in turn rel ies on several sources for its information , including the fol lowinq : o A Sa lmonel l a Surveillance System , ma intained by CDC s ince 1 9 6 3 , when several large outbreaks of salmonellosis were traced to commercial egg products . The purpose o f the surveillance system is to accumulate epidemioloqic data such as the age , sex , and county of res idence of pat ients from whom Salmonella isolates are submitted to state health departments for serotyping . Data are a l so kept on isolates from food and animals . o Investigations of outbreaks by state , local , and federal agencies . o Special epidemiol og i c and l aboratory surveil lance in selected counties . In 19 7 9 - 1 9 8 0 and 1 9 8 4 - 1 9 8 5 , the health authorities i n a strati f ied sample of urban and rura l count ies were asked to submit all salmonella isol ates , together with detailed epidemioloqic information , for a l l patients from whom isolates were obta ined . 12 The communities were chosen to provide about 5 % of the expected number of reported i solates . 2 4 Stra ins from known outbreaks were excluded . The isolates col lected in this way were obta ined throuqh the usual CDC reporting channels with no spec i fic e fforts at case- f inding ( R . Tauxe , CDC , 1 9 8 8 , personal communication ) . The isolates were tested for susceptibil ity to antimicrob ial agents and sometimes for the ir plasmid DNA content . These findings have been used to provide information on the rates of antib iot ic res i stance , hospital i z ation for il lness , and morta l ity from sa lmonel los is . The salmonella surveillance system conducted by CDC has shown a fairly constant rise in the annual number and rate of reported cases of nontypho idal salmonellosis at least s ince 1 9 5 5 ( Fiqure VI I -2 ) . The reasons for thi s rise are not clear . However , the bel ief that the rise was not s imply the result of better case find inq is supported by the observation
127 that there was n o appreciable change in the reported rate o f shige l losi s over the same period . In the years from 1 9 8 2 through 19 8 6 , the number of cases of salmonellos i s reported per year ranged from 4 0 , 8 6 1 to 6 5 , 3 4 7 . 8 OVer 9 0 % of the reported isolates were from symptomat ic individual s ( P . Blake and R . Tauxe , CDC , 1 9 8 8 , personal communication ) . There is substant ial underreporting o f salmonel los i s . 1 1 Indeed , it has been estimated that i n several outbreaks a s few as 1% o f cases were reported . 2 I n a telephone survey conducted recently during a mass ive epidemic of salmonel los i s , it was found that only about 1 0 % of symptomatic infections were reported . 2 5 Attempts to determine the number of cases more precisely run into the problem of def ining exactly what is a case . Should one include as cases only symptomatic pat ients with infections? Does passage of one or two loose stools qual i fy? Must symptoms be severe enough to interrupt normal activities for at l east 2 4 hours? Must symptoms be severe enough to require medical attention? The numbers might vary by one or two orders o f magnitude , and no answer i s inherently correct . Each investigator in the f ield must develop a conceptual de f inition that i s meaningful and useful for a speci fic study . Thi s must then be translated into operational terms : How can one col lect and interpret data so as to estimate both the number of cases by thi s definition and the degree of error l ikely to attend the estimate? An extens ive analysi s using three independent methods to derive these estimates for the annual incidence of salmonel los is produced estimates ranging from 8 0 0 , 0 0 0 to 3 , 7 0 0 , 0 0 0 infections . 1 1 A mean estimate of 1 . 9 m i l l ion infections in 1 9 8 4 would imply that about 2 . 5 % of infections ( about 5 0 , 0 0 0 cases ) had been reported to the C Dc . 1 1 Therefore , a n annual incidence of 5 0 , 0 0 0 reported cases o f nontypho idal salmonellosis i n the United States is a highly conservative estimate . A more probable f iqure is on the order of 8 0 0 , 0 0 0 cases per year , and the upper l imit could be as h igh as 3 , 7 0 0 , 0 0 0 cases per year . 1 1 Because many of the estimates cons idered critical for use in the model were based on data derived by the CDC from cases reported to that agency , the committee used the number of isolates reported as the starting point for the model ( see " Limitations in the Data , " above ) . In the estimat ion o f risk , the committee used the following f igures for low , midÂ range , and h igh-estimates for the number of reported cases o f salmonel losis per year ( U . S . only ) : 4 0 , 0 0 0 , 5 0 , 0 0 0 and 6 5 , 0 0 0 , respect ively . ANtiBIOTI C BES ISTAMCE OF SALMONELLA The proport ion of salmonella isolates from humans with
128 res istance to a t least one antimicrobial aqent was 1 6 % in the 1 9 7 9 - 1 9 8 0 CDC survey and 2 4 % in the 1 9 8 4 - 1 9 8 5 survey . 2 0 The proportion with resistance to two or more druqs , i . e . , multiresistant strains , increased from 1 2 % to 1 5 % durinq these same years . 2 0 These surveys avoided the countinq o f multiple isolates from the same outbreak o r the same patient . Chanqes in drug-res istance rates vary amonq the di f ferent salmonella serotypes . In the 1 9 7 9 - 1 9 8 0 study , the rate of res istance was hiqh for ï¿½ . heidelberg , but low for ï¿½ . typh imur ium ; in the 1 9 8 4 - 1 9 8 5 study , the rate o f res istance of ï¿½ . heide bberq decreased , and that of ï¿½ . typhimurium increased . 2 There were a l so chanqes in resistance to d i f ferent ant ibiotics . The rate of res istance to ampici l l in rose from 8 % to 9 % between the two study periods , and the rate o f res istance t o tetracycl ine rose from 8 . 6% t o 1 3 % ; by contrast , the rates of res istance to chloramphenicol and trimethoprim-sul famethoxazole were each 2% or less for both study periods . 2 0 At the request of this committee , the CDC provided additiona l informat ion which a l lowed the committee to calculate that 19 of 4 8 5 stra ins ( 3 . 9 % ) were res istant to both amp ic il l in and tetracycl ine , whether or not they were also res istant to other aqents . In a collect ion of 2 , 8 2 6 stra ins isol ated from humans from Massachusetts durinq 1 9 7 9 - 1 9 8 0 the preva lence of res istances was as follows : ampicil l in , 5 . 1 % ; tetracycl ine , 8 . 7 % ; tetracycl ine and ampic i l l in , 3 . 3 % ; tetracycl ine or ampici l l in , 1 0 . 3 % . The overa l l prevalence of res istant stra ins in the Massachusetts col lection ( see Table V- 1 ) was low , compared to other sources , perhaps because of the hiqh proport ion of qeneral l y susceptible ï¿½ . enteritidis ( 3 1% ) . For the est imation of risk , the committee chose the rates for occurrence of ant ibiot ic res i stance shown in Tab l e VI I-2 . The high est imates for res istance were chosen to account for the apparent increase in res istance rates over time , because the reported rates may underestimate the current prevalence of res istance . MORTALITY BATE FOR INFECTION BY RESISTANT STRAINS OF SALMONELLA In concept , the number of deaths from salmonella infections should be only the number of infected persons who died , and who would not have died in the absence of these infections . Thi s concept encounters serious problems in appl ication , because the causes of some deaths are d i f ficult to determine in ordinary conditions of med ical practice , and the asserted cause of death on a death cert i f icate often is unproved . Some di seases or conditions act j o intly to cause death ,
129 TABLE VI I - 2 RES ISTANCE O F SALMONELLAE TO ANTIMICROBIALS Rate of res istance of ww Mid-Ranqe Hiqh salmonellae to ; Estimate Est imate Estimate At least one antimicrobial 16% 24% 3 1% At least penici l l in/ampici l l in 10% 15% 2 0% or tetracycl ine source ; Adapted by the committee , from data in Table V- 3 . althouqh neither alone has resulted in death ; in such a case , what i s the underlyinq , " cause of death " ? I n some instances , a severe infectious disease is almost incidental to a severe underlyinq terminal condition ; sha l l we count such a death i f the salmone l l a infect ion only advances the time of death by a month , a day , or an hour? A deci s ion is needed about whether to count as a salmonella death the death of a person who was severely deb i l itated from other causes but whose uncontrol led salmonella infect ion contributed to the death . The committee recoqni z es that the u . s . National Center for Heal th Statistics , l ike other offices of vital stat i st ics , has consistent and wel l -developed rules for decidinq how to report these types of deaths . The committee understands the need for wel l -defined and consistent statistical data , especial ly for identi fyinq d i f ferences amonq populations and chanqes over t ime . However , the committee emphas i zes that these rules deal with the underlyinq conceptual problems in cons istent reportinq without solvinq them in a way that i s useful here . Table VI I - 3 summa ri zes recent data on reported death rates of patients with salmonel losis . The CDC study of 19 7 9 - 19 8 0 identi f ied seven deaths amonq 5 0 3 pat ients with nontyphoidal salmonellos i s , for a death rate of 1 . 4 % ; there was a s imilar rate in the 1 9 8 4 - 1 9 8 5 survey . 12 However , these were all deaths amonq pat ients reported to have salmonel los is , reqardless of speci fic causes of death . The committee asked for additional information reqardinq the role of salmonellosis in caus inq death . CDC was not able to provide such data for pat ients who died in the 1 9 7 9 - 19 8 0 study , but did provide additional information on the 8 deaths amonq about 6 0 0 patients included in the 19 8 4 - 1 9 8 5 survey . Accordinq to the committee ' s interpretation of those data , salmonellosis played an unknown role in three of the deaths and no role in four deaths , and it clearly contributed to the
TABLE VI I - 3 RECENT MORTALITY RATES FROM NON-TYPHOI DAL SALMONELLOS IS S pec i f i ed/or C omb i ned P r o s pec t i ve C a..u n i t y - Acqu i r ed c o..un l t y Nosoc i â¢ i a l CDC Infect i on N o s ocoâ¢ t al I nfect ion Soyr c e of Infect i on S u rve i l I Ince Mul t lÂ· Not Mul t i Â· Not Mu l t i Â· Not pa t .ll S u s c eD Res 12!Â£ Suscep R!l le!Â£_ Susc ep Rea Spec x.:CO!!!ï¿½eï¿½:::::nï¿½t<&l-- 1 I I I I I I 1 979 Â· 1 980 I I I I I I f 1 . 4X ! R o l e of S a l â¢one l l os t s i n caus i ng de a t h s I I I I I I 1 < 7t 503 > ! no t 1pec i t i ed . 1 984 Â· 1 985 I I I I I I j 1 . 4X ! On l y one dea t h c l ea r l y 1 t t r i bu t ab l e t o I I I I I I 1 < 81 600 ) I S i l mone l l o s h < s ee t e x t ) . 1 984 - 1 985 I I I I I I I I Io- 21 I < r eca l cu l at ed > ! I 1 1 I I I I 1 < 1 160 0 > I I I I I I I I I I I I I I I I I I I I I Outbreaks I I I I I I I I I I I I I I I I I I I I 1-' w us ou t b r e a k s ! I I I I I 1 0 . 21 f 4 . 21 I I 0 1 971 - 1 983 1 8 1 1 I I I I 1 < 4t 1 9 1 2 > 1 < 1 3 t 3 1 2 > 1 I I I I I I I I I I I US Outbreaks i 0 - 21 1 3 . 41 I 1 1 . 01 f 1 1 . TX I I I I f Molt of t h e da t i base w a s t h e 11me 1 9 7 1 - 1 98o 1 7 1 < 3! 1 32 1 > 1 < 71 2 05 > 1 1 < 2t20 2 > I < 30t256 > 1 I I I I â¢â¢ for 1 971 Â· 1983 . 1 8 I I I I I I I I I I N . E . US A 3 0 I 0 . 5X I I I I I I I I I ï¿½ ent e r t d ta , p r esu.1b l e drug - suscept i b l e 1 < 1 1 /2 1 19> 1 I I I I I I I l < not l t e t ed ) v i e g r1de A eggi I n I I I I I I I I I I N o r t h e a s t e rn USA . I I I I I I I I I I U n i t ed I I I I I I I I f 0 . 3X ! S ee Ch 1pter 6 f o r d i sc u t s i on of c a ses . IC i n g d oâ¢ I I I I I I I I I < 40t 1 2 , ooo > l I I I I I I I I I I M i dw e s t I 10. 11 I I I I I I I 1 1 4 deaths pr obab l y or po s s i b l y r e l a t ed t o US A 2 5 I 1 < 1 4 / 1 2 , 624 > 1 I I I I I I I S a l â¢one l l o s i t ; L. typh i mu r i UIII I n I I I I I I I I I l pa a t eu r hed â¢ I l k r es i s t a n t t o amp i c i l l i n , I I I I I I I I I I t h e t e t rlc yc ! l nes , c1 rben i e l l I i n , and I I I I I I I I I l s u l f i soxa z a l e . I I I I I I I I I I 8 Source : Adapted by the committe e , from d a t a by Cohen a nd T a uxe , 1 2 Holmb e rg , 1 7 , 1 Ryan , 2 5 and st . Lou i s . 3 0 u . s . outbr eaks 1 9 7 1 - 1 9 8 0 1 7 i n c l udes Pue rto R i co .
13 1 death o f only one patient , in whom the organi sm was i solated from blood ( P . Blake and R . Tauxe , CDC , 1 9 8 8 , personal communication ) . Accordingly , in only 1 patient o f about 6 0 0 ( 0 . 2 \ ) d i d salmonellosis clearly contribute t o the death o f the patient , although the rate may have been as high as 4 i n 600 ( 0 . 7\ ) . In a CDC review of outbreaks of salmonellos i s by Holmberg et al . , 17 encompassing the period 1 9 7 1 - 19 8 3 , the overal l death rate from drug-susceptible strains was 0 . 2 \ and from drug-res istant strains was 4 . 2 % . O f the 1 3 deaths caused by multiresistant stra ins , 8 were in elderly persons in the community , and 5 occurred among 18 infants in a s ingle hospital nursery . 18 The bas i s on which it was determined that salmonella infection caused or contributed to death was not stated . More recently , Holmberg et al . 1 7 reviewed both nosocomial and community-based outbreaks that were investigated by CDC and that occurred in the United States between 1 9 7 1 and 1 9 8 0 . These outbreaks were caused by various species of bacteria . Al l but one of the outbreaks of salmonellosis reported in thi s review had already been reported in the earl ier paper by this group . 1 8 In 1 0 o f the community-based outbreaks of salmonellosis that were ident i f ied as being caused by drug-susceptible stra ins , salmonel lae caused the death of three persons among 1 , 3 2 1 persons infected , or 0 . 2 % . By contrast , the death rate in four community-based outbreaks caused by multires i stant strains was 7 of 2 05 , or 3 . 4 % . 8 The death rate in seven o f the nosocomial outbreaks that were caused by drugÂ susceptible salmonel lae was 1 . 0% ( 2 of 2 02 pat ients ) but in nine o f the nosocomial outbreaks caused by multiple-drugÂ res istant salmonellae , the death rate was 1 1 . 7 % ( 3 0 of 2 5 6 patients ) . In both the community acquired nosocomial outbreaks j ust cited 17 , 18 the data do not allow comparison o f the ages of the individuals with salmonel los i s . Thi s comparison might be important in view of the greater number of deaths due to salmonella infections reported at the two extremes of age ( see Table VI I - 1 ) . Among the nosocomial outbreaks reported 17 , 1 1 of the 3 0 deaths due to multires istant salmonellae occurred in patients in neonatal intens ive care units , whereas the 2 deaths due to antimicrob ial-susceptible stra ins occurred in general hospital wards in patients whose ages were not spec i f ied . The committee consulted the CDC for additional detai l s on this issue but no further information was ava ilable ( S . D . Holmberg , 1 9 8 8 , personal communication ) . Thus , i f more outbreaks due to antimicrobial -resistant strains had involved infants and the elderly than outbreaks due to susceptible stra ins , the higher mortal ity associated with res istant strains might reflect such a d i fference in the population at risk .
132 A review of 6 5 outbreaks o f ï¿½ - enteritidis infection that occurred between January 1 9 8 5 and May 1 9 8 7 in the northeastern United States showed a death rate of 0 . 5 % ( 1 1 deaths among at least 2 , 1 1 9 cases ) . 3 0 Ten of these deaths occurred among 1 3 0 residents in nursing homes . Although ant imicrobial susceptibi l ity data were not given , strains of ï¿½ enteritidis nearly always have been found to be susceptible to commonly used ant ibiotics . Grade A eggs or foods containing eggs were impl icated in 7 7 % of the outbreaks in which a source of the infecting bacteria could be found . 3 0 The two papers on CDC ' s surveys o f outbreaks are summa ri z ed by Holmberg et al . 17 , 18 ( Table VI I - 3 ) and show a higher death rate from infection due to mult iple-drugÂ res i stant stra ins than from infection due to susceptible stra ins of salmonel lae . The more recent paper summa ri z ed data from outbreaks of infections that included bacteria other than salmonel lae and reported nosocomial and communityÂ acquired infections separately and more clearly than the earl ier summa ry . The authors concluded that for both community-acquired and nosocomial infections , the mortal ity rate , the l ikel ihood of hosp ita l i z ation , and the length of a hospital stay were usual ly at least twice as great for pat ients infected with drug-resistant strains as for those infected with suscept ible stra ins of the same species . 17 A higher mortal ity rate due to infect ion with drug-resistant stra ins of salmonel lae could result from : ( a ) a greater virulence of the resi stant strains , ( b ) a propensity for the drug-resistant stra ins to infect patients with dimini shed host defenses ( see "et i ologic fraction " below ) , or ( c ) the inefficacy of treatment with drugs to which the bacteria are res istant . Salmonella deaths are a l so reported by the u . s . National Center for Health Stat istics ( NCHS ) in the National Death Index ( NDI ) , which col lects data from the total u . s . and tabulates causes of death us ing the International Class i ficat ion of Disease ( ICD) categories on the death certificates . The data summari zed in the NDI do not provide any information on the individual cases , their locat ion , the infecting bacterial serotypes ( except that nontyphoidal salmonella infect ions are ident i f i ed as such ) , serotypes or their drug susceptibil ities . Therefore , these data do not have the same application for the analys i s of risk as the CDC data . It is useful to compare the salmonella deaths reported by both sources as a check on the accuracy of salmonella deaths reported by each . The NDI data are based on information given on the death cert i f icates completed by phys icians , coroners , or medical examiners . Summaries of deaths from salmonel los i s are given in Table VI I -4 for the years 19 8 0 to 1 9 8 5 . During this period , " other ( nontypho idal ) salmonellos i s " ( ICD-9 , code 0 0 3 ) was reported as the underlying cause of death in 8 2 - 1 17 deaths per year . In addition , during this same period ,
133 TABLE VI I-4 NUMBER OF DEATHS DUE TO SALMONELLOS IS ( ICD-9 , NO . 0 0 3 ) TABULATED IN THE NATIONAL DEATH INDEX ï¿½ ï¿½Â· lA * * 1985 117 2 18 1984 89 164 1983 82 154 1982 87 176 1981 104 1980 88 175 Source : National center for Health Stat i stics . 22 * uc â¢ Underlying cause of death ( i . e . ) disease was l isted as initiating cause of death on death cert i ficate ) . * * EA â¢ Entity axi s ( i . e . , d i sease was l i sted as contributing cause of death ) . salmonel los i s , whether it was only contributory or was thedirect cause of death , was mentioned on the death cert i f i cate in 1 5 4 -2 1 8 deaths per year . 2 2 These numbers of deaths , derived from sources and methods that are d i f ferent from those used by CDC , are within the range of death rates cited by the CDC . For example , if salmonellosis was the underlying cause of death in 1 0 0 deaths per year ( intermediate between 8 2 and 1 17 deaths per year ) and these deaths are cons idered to occur among the 5 0 , 0 0 0 reported cases of salmonellosis per year , the death rate would be 0 . 2% . For the approximately 2 0 0 cases per year in which salmonellos i s was at least a contribut ing factor , the death rate would be 0 . 4 % . The NDI data may underestimate the number of deaths from salmonellosis for several reasons : ( 1 ) a death certi ficate may be f i l led out by a physician or other authori z ed person who is not ful ly fami l iar with the il lness of the deceased and may not recognize the contribution of the infection to " card iac arrest " or some s imilar non-infectious process ; ( 2 ) a salmonella infection may be recogni z ed and reported as septicemia , meningitis , or other infection without identi fication of the causative organi sm ; or ( 3 ) fa i lure to take a bacterial culture from a patient with salmonellosis as a complication of a terminal i l lness might lead to fa ilure to identi fy the organism .
134 The data on case studies i n Table VI I - 3 suqqest that the death rate may ranqe from 0 . 2 % to O . S% in community-based salmonel l a infections caused by druq-susceptible strains , but could be as hiqh as 1 . 4 % i f the CDC surve i l l ance data were for susceptible stra ins , a fact which cannot be determined from the reports . The death rate may ranqe from 0 . 1% to 3 . 4 % or 4 . 2 % in community-based outbreaks caused by mult iple-druqÂ res istant stra ins ; however , the two h iqher values for death rates from resistant stra ins are derived from on overlappinq body of data and are not independent . The death rate was 1% in nonsociomal outbreaks caused by susceptibl e stra ins and was as hiqh as 1 1 . 7 % in nosocomial outbreaks caused by multiple-druq-resistant strains . The death rates used in the risk assessment model are summa ri z ed in Table VI I - S . The committee considered whether the low est imate for strains resistant to at least one druq should be set at 0 . 1 % in accord with the report by Ryan et a1 . 2 s on a very larqe outbreak of milk-borne salmonel losis . The committee concluded that the reported death rate in that outbreak was unusual ly low , perhaps because extens ive publ icity led to substantially above-averaqe reportinq of marqinal cases ( the denominator of the reported death rate ) . Each outbreak is thus cons idered a sample of one and not weiqhted accordinq to the number of persons a ffected . No death rates were ava ilable from the l iterature for strains res istant speci fically to at least penicil l in/ ampici l l in or the tetracycl ines , althouqh res istant stra ins isolated in epidemics were frequently res istant to at least one of these druqs . Thus , the committee elected to use the same death rate of 0 . 2 % for strains res istant to penic i l l in/ampicil l in or the tetracycl ines as to any other druq ( see Table VI I - S ) . For the mid-ranqe and hiqh estimates of the death rate from stra ins with no res istance , the committee elected to use estimates that spanned the ranqe of values shown in Table VI I - S for stra ins of thi s kind , whether the infections were community-acquired or nosocomial . For strains with druq res istance , the committee bel ieved that the plaus ible mid-ranqe or hiqh death rates were appreciably hiqher than those for stra ins with no res istance , for three reasons : ( a ) some patients , althouqh probably few , would receive the "wronq" druq intentiona l ly or inadvertently ; ( b ) druq-res istant stra ins would tend to cause some infections in the " etioloqic fraction " cateqory ( d iscussed bel ow) in patients who miqht be debil itated and more susceptible to the consequences of infection ; ( c ) the data of Holmberq et al . summari zed in Table VI I - 3 indicate a hiqher death rate from res istant than from susceptible stra ins . As d iscussed earl ier , there are theoretical reasons why druq-resistant strains miqht be more virulent than druqÂ susceptible stra ins , includ inq the pos s ibil ity that the res istance plasmids have acqu ired virulence qenes ( or that
135 TABLE VI I - 5 RANGE O F DEATH RATES (AS A PERCENTAGE ) FOR SALMONELLOS IS FROM SUSCEPTIBLE AND DRUG-RES ISTANT STRAINS Jii&timsatt=& ï¿½ Mid-Rsanqe High susceptibl e 0.2 0.5 1.0 Res istant to 0.2 1.0 4.0 at least one druq Res istant to at least 0.2 1.0 4.0 penici l l in/ampicil l in or tetracycl ine Source : Prepared by the committee ( see Table VI I I - 1 ) . virulence plasmids have acquired res istance qenes ) . It miqht b e arqued that the death rate for strains res istant t o atleast penici l l in/ampicil l in o r the tetracycl ines should be hiqher than for strains res istant to any druq , because the former would be more l ikely than the l atter to l ead to problems with the "wronq" choice of druq , or with the " etioloqic fraction . " However , l ackinq any spec i f ic data on thi s point , the committee chose to use the same mid-ranqe and hiqh estimates for these two kinds of strains . In any event , it seems unl ikely that the results from the risk model wil l be much influenced by the use of s imilar values for these two kinds of stra ins . The reported death rates for infection by res istant strains in Table VI I - 3 are for multiresistant stra ins , whereas the des iqnation in Table VI I - 5 is for strains res istant to at l east one aqent : the committee cons iders thi s distinction to be of minor consequence for the present analys i s . FRACTION A8SOCIATJ:iD WITH FARM QRI GIN A critical step in the risk estimate is the determination of the source of res istant strains o f salmonel lae that cause infection in humans . In fact , the true " oriqin" o f a stra in of salmonel lae that causes infection in humans--i . e . , whether it arose from a food product , contact with another person or a pet , or some other source--is almost never known , except in outbreaks that are investiqated . However , there is a common bel ief that for most strains of nontyphoidal salmonel lae , the proximate
13 6 source i s usua l ly food animals or food products derived from anima l s . This bel ie f is supported by the f indings that carriage of nontyphoidal salmonel lae in humans is general ly brie f and that a wide variety o f commonly consumed foods are often contaminated by stra ins of salmonel l ae and by the evidence ( summa ri z ed below) that , in most outbreaks in which the source could be traced , the source has been food products originating on the farm . Rel iance on the results derived from the analys i s o f outbreaks i s problematical , because salmone l l a infections in humans are usual ly sporadic . Of course , i f adequately detai l ed investigation could be done , it is l ikely that many " sporadic" cases would turn out to be smal l epidemics . There is no evidence that salmone l l a stra ins isolated from outbreaks are dist inct from stra ins isolated from sporadic cases : the epidemiologic e ffects o f a given strain are presumably related to the number of organisms in the infecting inoculum and the number of people exposed to thi s inoculum . Because of these cons iderations , the committee concluded that the data from outbreaks are usable in the present context . In CDC ' s review of 52 outbreaks of salmonel losis , food anima l s or their food products were impl icated in 11 of 16 outbreaks ( 69 % ) caused by drugÂ res istant salmone l l ae , 6 of 16 ( 3 8 % ) outbreaks caused by drug-susceptible salmone l l ae , and in 1 of 9 outbreaks ( 1 1 % ) caused by salmonel lae o f unknown susceptibil ity . 18 By consensus , the committee concluded that the low estimate for the l ikely percentage of res istant stra ins that originated in farm anima l s or the ir products was 5 0 % , and the upper l imit of this estimate might be as high as 1 0 0 % . For the mid-range estimate , 7 0 % was used--a value s imilar to that found in the CDC ' s review of outbreaks cited above . FRACTION OF ANTI BIOTIC-BES I STANT STBAINS CAUSED BY SUBTHEBAPEUTIC USE OF ANTIBIOTI CS Because of the pauc ity of data , the most uncertain aspect of the committee ' s risk analys is is the estimation of the portion of drug res istance in salmonel l ae o f farm origin that i s attributabl e to the subtherapeutic use of antibi otics . Farmers use antimicrobials in subtherapeutic dosages in feed for two purposes : ( a ) growth promotion , and ( b ) prophylaxis ( such as the prevention of atrophic rhiniti s in swine o r " shipp ing fever complex" in cattle ) . FDA ' s definition of subtherapeut ic use includes use for both growth promotion and prophylaxis . The committee ' s obj ective is to devel op data for this combined use of antimicrob i al s , as well as for use in growth promotion alone . The committee could find no data that bear directly on the relative contributions to the development of drug
137 res istance caused by any of the three maj or dosaqe reqimens for antimicrobials--for therapy , for qrowth promotion , or for prophylaxis . Indeed , the data are l imited and inconclus ive reqardinq the relative contribution of chronic low-dose administration vs . intermittent hiqh-dose admin istration of antimicrobial aqents in fosterinq druq res istance ( see Chapter I I I ) . OVeral l , the l imited data ava i l able to the committee suqqest that chronic exposure to low concentrations o f antimicrobials is at l east as l ikely to foster res i stance as intermittent exposure to hiqh concentrations . Given the substantial uncertainties about the causa l relationship between the type of antimicrobial dosaqe reqimens and development of res istance , the committee adopted as its mid-ranqe estimate of the relative contributions of the three maj or farm uses of ant ibiotics the approximate proportions ( percentaqes by weiqht ) of druqs administered nationwide for each of these purposes to animals ( see Chapter IV) . Even this seeminqly straiqhtforward approach proved di f ficul t , because of the problems in obta ininq rel iabl e estimates o f the amounts of the various antibiotics used for each of the three ma in purposes ( see Chapter IV) . As a startinq point , it appears that , overall , about 1 2 % o f ant ib iotics s o l d for veterinary u s e is used for therapeutic purposes and about 8 8 % is used in subtherapeutic dosaqe reqimens ( see Table IV-9 for the source of these percentaqes ) . Based on the tonnaqe ratios shown in Table IV-9 , some two-thirds of the druqs used for subtherapeutic purposes is qiven for prophylaxis and one-third qrowth promotion ; this would result in a partition of the 88% into about 60% for prophylaxis and 2 8 % for qrowth promotion . However , in the j udqment of committee members the fraction used for prophylaxis is probably about three-fourths . Thi s would result in a part ition of the 8 8 % used subtherapeutica l ly into 6 6 % for prophylaxis and 2 2 % for qrowth promotion . The arithmetic means of these estimated percentaqes are 6 3 % ( 6 0 - 6 6 % ) for prophylaxis and 2 5% ( 2 2 -2 8 % } for qrowth promotion , as shown in Table VI I - 6 . Accordinqly , on the s impl e assumption that the contribution of any druq used o n the farm to the development of druq res istance would be in l inear proport ion to the amount used for each of the three purposes , the committee chose a mid-ranqe estimate of 1 2 % for the contribution of ï¿½ drugs used for therapeutic purposes and a mid-ranqe estimate of 8 8 % for the contribution of ï¿½ drugs used subtherapeutical ly ( 6 3 % for prophylaxis and 2 5 % for qrowth promotion ) . The committee recoqn i z es that there may not be a l inear rel ationship between selection of ant ibiotic res istance and the total amounts of antibiotics used in farm anima l s . By consensus , the committee chose the plausible low estimate for
138 TABLE VI I - 6 ESTIMATED PERCENTAGE O F ANTI BIOTIC RESISTANCE I N STRAINS O F FARM ORIGIN CAUSED BY SUBTHERAPEUTIC USE OF ANTI BIOTICS IN ANIMAL FEED Contribution of growthÂ Proportion of promotion use to resistance Tonnage for Low Mid-Range High Use Specified Use Estimate Estimate Estimate AnY Resistance caused by AnY Drua Therapeutic 12 15 12 8 Subtherapeutic 1 1 } Prophylaxis 60-66 80 63 42 ( 63 ) 85 88 92 Growth 2 2 -2 8 5 25 50 promotion (25) 100 100 100 100 Est imated Percentage o f Pen ic i l l i nLAmQi c i l l in Or Tetracyc l i n e Re s i stance Caused by Administration o f Pen i c i l l i nLAmQi c i l l i n o r the Tetracyc l i n e s Therapeutic 10 14 10 6 Subtherapeutic Prophylaxis 60 81 } 86 60 \ 90 34 } 94 Growth 30 5 30 60 promotion 100 100 100 100 Source : Table prepared by the committee . - ---
139 AnY gxyg g iven for growth promotion of 5 % and the high est imate of S O % . These estimates were chosen bearing in mindthat the most plausible figure for the actual tonnage (mid-range estimate ) of drugs used for growth promotion was 2 5 % . The remaining percentages of AnY drugs used therapeutica l ly or prophylactica l ly were partitioned with the same ratio as for the mid-range estimates-- i . e . , about Â· 1 : 5 -Â to fill out the data base for the low and high estimates . The l ower hal f of Table VI I - 6 presents the committee ' s estimates o f the contribution of penicillin/ampicillin or the tetracyclines to the development of drug res istance when used in the feed of farm anima l s for one of the three maj ortreatment purposes . These were derived in an anal ogous manner . However , for these antibiotics the committee assumed that a higher percentage was used for growth promotion than shown for AnY gxyg in the upper hal f of Table VI I - 6 . In part icular , it was cons idered that the addition of the tetracycl ines to swine feed was predominantly for the purpose of growth promotion ; therefore , the mid-range estimate of 2 5 % for the contribution o f drugs given for growth promotion was raised to 3 0% for stra ins res istant to penic i l l in/ampici l l in or the tetracycl ines . The other two values in the mid-range estimate for penicill in/ ampici l l in or the tetracycl ines were reduced as fol l ows : for prophylaxi s , 6 3 % to 6 0 % ; and for therapy , 1 2 % to 1 0 % . The ratio of therapeutic to prophylactic use thus became 1 : 6 , which corresponds to the proportional amounts given to animals for these purposes . Thus , by committee consensus , 5 % was chosen for the ï¿½ estimate and 6 0 % for the high estimate of the proportional use o f antibiot ics for growth promotion . Thes e estimates are somewhat higher than those for res istance to AnY ï¿½ in l ight o f the extens ive use of the tetracycl ines for growth promotion . The rema ining estimates for therapeutic and prophylactic use were partitioned in a ratio of about 1 : 6 . Thus , a l l estimates have been adj usted sl ightly so that the total for each of the three types of use in each column is 1 0 0 % . This assumes that essential ly all the antibiotic res istance found in salmonel lae encountered on the farm is rel ated to the amount of antimicrobial drug in the aggregate , used for each of the three maj or types of appl ication in the aggregate . PREYENTABLE CASES OF SAiliQNELLOS IS : "ETIOLOGIC fRACTION " The approach used in the risk model has been to estimate the number of persons who die each year as a result of infection with drug-resistant strains of salmonel l ae that originated on the farm ( i . e . , salmonel lae isolates from meat or animal food products , eggs , or milk) and for which the drug resistance was selected by the subtherapeut ic doses of
14 0 ant ibiotic drug in animal feed . However , at the beginning o f this sect ion the committee acknowledged that it cannot estimate the total number of cases of salmonel losis , the pro f i l e of drug susceptib i l ities , or the source of the bacterial stra in that would l ikely occur in the United States if subtherapeutic doses of any antibiotics , or of peni c i l l in/ ampici l l in or the tetracycl ines , were not administered to farm anima l s . I t might be argued that , i f a l l subtherapeutic use o f penici l l in/ampici l l in o r the tetracycl ines were stopped , deaths due to in fection by stra ins of drug-resistant salmonel lae would be replaced by a l ike number of deaths caused by drug-susceptible salmone l l ae . However , there are at l east three ways in which drug res istance itsel f might contribute to a higher total number of deaths from salmonel l a infection : ( a ) by l eading t o a "wrong " choice of drug for treatment ( presumably an uncommon event ) , ( b ) by being intrins ica l ly more virulent and hence lethal ( a poss ibi l ity for which there is some evidence , but which might in turn relate to wrong choice of drug or to etiologic fraction ) , or ( c ) by caus ing some infections that would not have occurred but for the resi stance of the in fecting bacterial stra in to the antimicrobials administered . Infections caused in the third way have been cal l ed the " etiologic fraction , " and the resulting cases of i l l ness are termed " excess cases . " The concept of the " etiologic fraction" arose from two observations : first , the ingestion by salmone l l a carriers o f drugs t o which the salmonel lae were res istant in occas i onal instances appeared to provoke the development of cl inical salmonel losis : 1 2 , 2 3 second , in one outbreak of salmonel los i s , the association between ingestion of penicil l in or other antimicrobials and the devel opment of infection was so striking that it led to the initial suspicion that the drug was contaminated with sa lmone l l ae . 1 6 Subsequently , contro l l ed studies have documented repeatedly that antimicrobial ingestion does enhance the l ikel ihood o f infection by drug-resistant salmonel lae in epidemic situations . The hypothesis for which supporting evidence ï¿½ exists in anima l s 5a ï¿½ S b , 2 0a , O b , 2 0c is that the antimicrobial drug suppresses the drug susceptibl e competing fecal flora , and enhances the opportunity for the pathogen to become implanted or , in carriers , to prol i ferate and cause d isease . The e ffect is to lower the s i z e of the inoculum needed to cause infection . For purposes of calculation , the etiologic fraction is determined by multiplying the relative strength of associat ion between the recent ingestion of an ant imicrobial agent and the l ikel ihood of devel opment of salmone l l osis-- i . e . , the " odds ratio " --by the proport ion of patients with that risk factor . In a recent review by Cohen and Tauxe 1 2 of various outbreaks , the proport ion of cases in the " etiologic fraction" ranged from 16 to 6 4 % . Whether
14 1 pat ients in the " etioloqic fraction" are more susceptibl e to infect ion with salmonel lae and whether they are more l ikely to die of thi s infection is not known , i . e . , the rel at ive death rates in the " etioloqic fraction" are not known . As the proportion of bacterial stra ins that are druq-res istant increases , cases belonqinq to the " etioloqic fract ion" should constitute a l arqer and larqer proportion of all infections . Tab l e VI I - 7 summariz es quantitative evidence from six recent studies about the increased risk of druq-resistant salmonel losis amonq individual s takinq antimicrobial aqents . As an i l lustration , the information from Adl er et a1 . 1 is summari z ed in Table VI I-8 where the fiqures in parentheses are obta ined by d i fference from the four fiqures in the top l ine of Table VI I-7 . The odds ratio is then calcul ated from the cross products of the four " internal " cel l s of Table VI I Â s a s fol l ows : OR = ( 2 8x19 ) I ( 2 1x8 ) = 3.2 Other l ines in Table VI I - 8 are to be interpreted s imilarly . The odds ratios , which are close estimates of rel at ive ri sks , are presented without statistical confidence l imits , because confidence l imits express only the uncertainty due to random error , and each of these sources is subj ect to cons iderab l e addit ional nonrandom uncertainty in any qenera l i z ation to a broader population . For example , a study of a s inql e stra in ( as in an outbreak) violates the basic assumption of statistical independence : controls drawn from patients on a pediatric ward may be both more vul nerabl e and more exposed to res istant stra ins than infants in qeneral ( or than the population as a whol e ) : and household contacts of pat ients may tend to share patterns of antib iot ic use with the patients . In addition , the sources of subj ects were not always wel l characterized , antibiotic res istance was determined in d i fferent ways , and methods o f ascerta ininq druq use varied . Neverthel ess , the odds ratios cons idered in thi s analysis are thouqht to be appl icable to patients who inqest peni c i l l in/ ampici l l in or the tetracycl ines and to salmonel lae that are res i stant to those druqs , in that they were derived primarily from studies in which those were the druqs involved in producinq the " etioloqic fract ion . " Given these d i fferent kinds of uncertainty , the cons istency in. the elevation of odds ratios is impress ive . However , because these odds ratios refer to d i f ferent sorts of subj ects and diseases ( i l l chi ldren , outbreaks vs . sporadic cases , etc . ) and because one may expect such di fferences to affect relat ive risks , none of these fiqures alone is ent irely suitable for estimatinq the odds ratio for the pub l i c at larqe or for estimatinq the part of the total national burden attributable to the overqrowth o f res istant stra ins when the normal flora is suppressed by antibiotic
142 TABLE VI I -7 SOME STUDIES OF THE FREQUENCY OF ANTI BACTERIAL THERAPY IN INDIVI DUALS INFECTED WITH ANTIMI CROBIAL-RES ISTANT SALMONELLA STRAINS Both Infected with Infected Senior Total Res istant Recent and Recent Odds Author_ l sub"iecta SA1111l'llftAl l AA IDrua Usa Drua UsA Batio Adler 76 patients 36 multi- 4 9 took 28 3.2 19701 on a pediatric resistant semisynthet ic ward ( s ingle penicil l in or stra in) ampic i l l in , l timA not stated Rileï¿½ ( a ) 5 0 4 cases , 66 resistant 4 3 took 1 or > 1 13 3.3 1984 4 geographically to 2 or >2 antimicrobials dispersed anti- within 4 weeks mi l'!,..nh i a 1 a ( b ) 4 3 pat ients 13 resistant 2 5 took 12 15 . 7 receiving strains* ampic i l l in , antimicrobials amoxici l l in , I or oenici l l i n Holmberg ( a ) 2 1 patients 10 res istant 7 took 7 Infinite 1984 16 with &. DAKRQDi ( s ingle amoxici l l in or infection stra in) penicill in within 1 week ( b ) Same 10 drug- Same Same 7 Infin ite res istant cases + 2 9 household ... .... _ .. . ..... .. ( c ) Same 1 0 drug- Same Same plus 7 I n f i n ite res istant cases + ophalorid ine 27 non-s Mac 4 8 5 isolates , 117 res istant 63 took ant i - 23 2.0 Donald geographically to at microbial a 198720 dispersed least one within I antibiotic 4 ... Spika 4 5 cases , 8 8 4 5 epidemic 13 took 11 13 . 9 198729 matched controls multi- penic i l l in or resistant tetracycl ine Is l within 1 month Ry an 50 cases , 50 5 0 epidemic Hot stated Hot stated 5.5 19 8 7 2 5 matched controls multi- resistant :s tvohimuriWD S ource : Adapted by the committee from Adler et a l . , l R i l ey , 2 4 Holmberg , 1 7 MacDona ld , 2 0 S pika , 2 9 and Ryan . 2 5 * Not c l ear whether res i stant to any ant ibiot i c , to 2 or more ant i b i otics , or to pen ic i l l ins (with or without other ant i b i o t i cs ) .
143 TABLE VII-8 NUMBERS OF PATIENTS WITH RESISTANT SAlMONELLA US ING ANTIMICROBIALS bliltAot SAlGDÂ§l.l.A liÂ§QÂ§Dt Qrua lllÂ§ All SubjgQts In_ lf2_ Yes 28 (8) 36 No .u.u .un 1.!.Ql Total s 49 (27) 76 S ource : Adapted from Adl er et al . 1 use . Those studies that come closest to ideal for estimating a population-wide odds ratio are those of Holmberg et al . , ï¿½ 6 , 1 8 Spika e t al . , 2 9 and Ryan e t al . 2 5 The three estimates are : infinite ( based on a sma l l sample ) , 1 3 . 9 , and 5. 5. The committee is incl ined to bel ieve that the true popul at ion-wide odds ratio , for cases s imilar to those regularly reported to CDC , for oral intake of any antimicrobial in common use among humans and for salmone l l ae res istant to that drug ( and perhaps others ) , may be about 5 and is probably between 2 and 2 0 . Clearly , these estimatesare uncerta in , and additional research is needed to improve them . A population-wide estimate of the proportion of salmone l l a infections or deaths ( the " etiologic fraction " ) can be derived from an odds ratio combined with an estimate of the proport ion of the population taking ant ib iotics as : EF = ( OR- 1 ) P/ [ 1+ ( OR- 1 ) P ] , where OR i s the odds ratio and P is the proport ion exposed to the risk factor ( here , antibiotic use ) . The definition o f "taking ant ibiotics" must b e close t o that used in estimating the odds ratio , so the committee reviewed the studies in Tabl e VI I-7 to see whether any could be used for this purpose . There are uncertainties in each of the studies , but the closest seem to be those of Holmberg ( none of 2 9 househol d contacts had taken ant ib iotics within one week , giving a rate of use of 0 . 0 % ) and Spika ( 2 of 8 8 matched controls had taken antibiotics within one month , or 2 . 3 % ) . Clearly , the population-wide figure is l arger than Holmberg ' s 0 . 0 % , and it may be close to Spika ' s figure , scaled down from
144 2 . 3 % for use i n the past month to , perhaps , 0 . 5 % per week . Thus , our best estimate here is 0 . 5 % use within the past week , to which we attach a high estimate of 1% and a l ow est imate of 0 . 2 % . Using the l ow , mid-range , and high estimates o f both OR and P , the committee produced nine estimates of the proportion of human salmonella infections due to this mechanism a l one ( the etiologic fraction ) : OOPS BATIO Antibiotic Use <Past Week> ï¿½ ï¿½ ï¿½ 0 . 2% 0 . 2% 0 . 8% 4% 0 . 5% 0 . 5% 2 . 0% 9% 1 . 0% 1 . 0% 4 . 0% 16% Thus , the committee ' s best estimate is that 2 % of salmonel l a infections of the sort reported to CDC are a di rect result of the use of ant ibiotics by persons who harbor salmonel lae res istant to those ant ibiotics . We think it unl ikely that the high estimates of both OR and P hold or that both l ow estimates hold ( the upper left and l ower right corners of the table ) . The range of the est imates excluding those two poss ibil ities is 0 . 5% to 9 % . The committee ful ly recognizes the great uncertainty of the est imates here . Nevertheless , we bel ieve that the estimates are worth present ing , partly because they embody our best j udgment about the matter , and partly because they point in a clear manner to a need for additional research . The uncerta inty i s inherent i n the l imitat ions o f ava ilable data , and other methods of analys is and presentation would s imply hide the uncertainty , rather than reduce it . BEFEBENCES 1. Adler , J . L . , R . L . Anderson , J . R . Boring , J rd , and A . J . Nahmias . A protracted hospital -associated outbreak of salmonellos i s due to a multiple-ant ibioticÂ res istant stra in of Salmonella indiana . J . Pediatrics 7 7 ( 6 ) : 9 7 0 - 9 7 5 , 19 7 0 .
145 2. Askeroff , B . , s . A . Schroeder , and P . s . Brachman . Salmone l l os i s in the United states--a f ive-year review . Am . J . Epidem . 9 2 : 1 3 -2 4 , 19 7 0 . 3. Blaser , M . J . , D . N . Taylor , and R . A . Feldman . Epidemiology of Campyl9bacter ieiuni infections . Epidemiol . Rev . 5 : 1 5 7 - 1 7 6 , 1 9 8 3 . 4. Blaser , M . J . Campyl obacter Species , pp . 1 2 2 1 - 12 2 6 . I n G . L Mande l l , R . G . Douglas , Jr . , and J . E . Bennett , Eds . Principles and Practice of I nfectious Diseases , Second Edition . New York : John Wiley and S ons , 1 9 8 5 . s. Blaser , M . J . , J . G . Wel l s , R . A . Feldman , R . A . Pol lard , and J . R . Al lan . The Col laborative Diarrheal Disease study Group : Campylobacter enteritis in the u . s . , A multicenter study . Ann . Intern . Med . 9 8 : 3 60 - 3 65 , 1983 . Sa . Bohnhof f , M . , B . L . Drake , and c . P . Mil l er . The effect of streptomycin on the susceptibi l ity of the intestinal tract to experimental salmone l l a infections . Proc . S oc . Exptl . Biol . & Med . 8 6 : 1 3 2 - 1 3 7 , 1 9 5 4 . Sb . Bohnhoff , M . , and c . P . Mil l er . Enhanced susceptibi l ity to salmonella infection in streptomycin treated mice . J . Inf . Dis . 1 1 1 : 1 17 , 1 9 6 2 . 6. Boyce , J . M . Yers inia Species , pp . 1 2 9 6 - 1 3 0 1 . In G . L . Mande l l , R . c . Douglas , Jr . , and J . E . Bennett , Eds . Princ ipl es and Practice of Infectious Disease , Second Edition . New York : John Wil ey and Sons , 1 9 8 5 . 7. Carter , A . o . , A . A . Borczyk , J . A . Carlson , B . Harvey , J . c . Hockin , and et al . A severe outbreak o f Escherichia Â£Qli 0157 : H7 -assoc iated hemorrhagic col itis in a nurs ing home . N. Engl . J . Med . 3 17 ( 2 4 ) : 1 4 9 6 - 1 5 0 0 , 1987 . 8. MacDonald , K . L . , and P . M . Gri ffin . Foodborne d isease outbreaks , annual summary , 1 9 8 2 . Morbidity and Mortal ity Weekly Report 3 5 ( 1SS ) : 7 , 19 8 6 . 9. Centers for Disease Control . PHLS Communicable Disease Surve i l l ance center : Gastrointestinal infections for 1977-1982 . Communicable Dis . Report p 1 , Jan . , 19 8 3 . 10 . Tauxe , R . v . , N . Hargrett-Bean , c . M . Patton , and I . K . Wachsmuth . Campylobacter isolates in the United states , 19 8 2 - 19 8 6 . Morbidity and Mortal ity Weekly Report . 3 7 ( SS -2 ) : 1- 1 4 , 1 9 8 8 .
146 . 11 . Chal ker , R . B . , and M . J . Blaser . A review o f human salmonel losis : I I I . Magnitude of salmone l l a infection in the United States . Rev . I nfect . Dis . 1 0 : 1 1 1 - 12 4 , 1988 . 12 . Cohen , M . L . , and R . v . Tauxe . Drug-res istant Salmonella in the United States : an epidemiologic perspective . Science 2 3 4 : 9 64 - 9 6 9 , 19 8 6 . 13 . Fagerberg , D . J . Bacterial Antimicrobial Res istance and FDA 2 1 CFR 558 . 1 5 Requirements With Speci f ic Results From Testing Oxytetracycl ine As A Feed Additive I n Beef cattle . Pfi zer ' s 3 6th Annual Research Conference , Nashv i l l e , Tennessee , May 4 , 1 9 8 8 . 14 . Finch , M . J . , and L . w . Ri ley . Campylobacter Infections in the u . s . Arch . Intern . Med . 1 4 4 : 1 6 1 0 - 1 6 1 2 , 1 9 8 4 . 15 . Food and Drug Administration . Epidemiologic study of campyl9bacter j eiuni Infections in Dubuque , I owa Between April 1 9 8 2 - March 1 9 8 3 . Contract Report 2 2 4 -8 1-7 09 3 , 1984 . 16 . Holmberg , s . D . , M . T . Osterhold , K . A . s inger , and M . L . Cohen . Drug-Res istant Salmonella from Animal s Fed Antimicrob ial s . N . Engl . J . Med . 3 1 1 ( 1 0 ) : 6 1 7 - 62 2 , 1 9 8 4 . 17 . Holmberg , s . D . , s . L . Solomon , and P . A . Bl ake . Health and economic impacts of antimicrobial res i stance . Rev . Infect . Dis . 9 : 1 0 6 5 - 1 07 8 , 19 8 7 . 18 . Holmberg , s . D . , J . R . Wel ls , and M . L . Cohen . Animal Â ta-man transmiss ion o f antimicrobial -resistant Salmonella : I nvestigat ions of u . s . outbreaks , 1 9 7 1 - 19 8 3 . Science 2 2 5 : 8 3 3 -8 3 5 , 1 9 8 4 . 19 . Johnson , K . E . , c . M . Nolan , and A . K . Ca in . CommunityÂ wide surve i l l ance of Campylobacter jejuni , p . 1 6 1 ( abstracted ) . I n Program and Abstracts o f the 2 2 nd Interscience Conference on Antimicrobial Agents and Chemotherapy . Washington , D . C . : American S ociety for Microb iol ogy , 1 9 8 2 . 20. MacDonald , K . L . , M . L . Cohen , N . T . Hargrett-Bean , J . G . Wel l s , N . D . Puhr , s . F . Col l in , and P . A . Blake . Changes in antimicrob ial res istance of Salmonella isolated from humans in the United States . J . Amer . Med . Assn . 2 58 : 1 4 9 6 - 1 4 9 9 , 19 8 7 .
147 2 0a . Meyne l l , G . G . Some Factors Affectinq the Res istance of Mice to Oral Infection by Salmonella typhimurium . Proc . R . S oc . Med . 4 8 : 9 1 6 , 1 9 5 5 . 2 0b . Meyne l l , G . G . , and T . v . Sabbaiah . Antibacterial mechani sms of the mouse qut . I . Kinetics o f infection by salmone l l a typhimurium in normal and streptomycinÂ treated mice studied with abortive transductants . Br . J . Exptl . Path . 4 4 : 1 9 7 -2 08 , 1 9 6 3 . 2 0c . Mil l er , c . P . , and M . Bohnhoff . Chanqes in the mouse ' s enteric microflora associated with enhanced susceptibi l ity to salmonella infection fol l owinq streptomycin treatment . J . Inf . Dis . 113 : 59-66 , 1963 . 21. National Research Council . The E f fects On Human Health of Subtherapeutic Use of Antimicrobials in Animal Feed . Washinqton , D . C . : National Academy Press , 19 8 0 . 22 . Nationa l Center for Health Statistics , Mortal ity Stat istics Branch , Publ ic Health Service . Vital Statistics o f the United States ( and National Death Index ) . Hyattsv i l le , Maryland . 2 2 a . Pa i , c . H . , R . Gordon , H . v . S im , and L . E . Bryan . Sporadic cases of Hemorrhaqic Col itis associated with ï¿½ ï¿½ 0 1 57 : H7 . Cl inical , epidemioloqic , and bacterioloq ic features . Ann . I nt . Med . 1 0 1 ( 6 ) : 7 3 8 - 742 , 1984 . 23 . Remis , R . s . , et al . Sporadic cases of hemorrhaqic col itis assoc iated with ï¿½ ï¿½ 0 1 57 : H7 . Ann . I nt . Med . 1 0 1 : 6 2 4 - 62 6 , 1 9 8 4 . 24 . Riley L . w . , M . L . Cohen , J . E . Sea l s , M . J . Blaser , K . A . Birkness , N . T . Harqrett , s . M . Martin , and R . A . Feldman . Importance of host factors in human salmone l l osis caused by multiresistant stra ins of Salmonel la . J . Infect . Dis . 1 4 9 : 8 7 8 - 8 8 3 , 1 9 8 4 . 25 . Ryan , c . A . , M . K . Nickels , N . T . Harqrett-Bean , M . E . Potter , T . Endo , et al . Mass ive outbreak of antimicrobial -resistant salmonel losis traced to pasteurized milk . J . Amer . Med . Assn . 2 58 ( 2 2 ) : 3 2 69 - 3 2 7 4 , 1987 .
148 . 26. Schmid , G . P . , R . E . Schae fer , B . D . Pilkayt i s , J . R . Schaefer , J . H . Bryner , L . A . Wintermeyer , and A . F . Kaufmann . A one-year study of endemic Campylobacterios is in a midwestern c ity : Association with consumption of raw milk . J . Inf . Dis . 1 5 6 : 2 1 8 -2 2 2 , 1987 . 27 . Seattle-Kinq County Department o f Publ ic Health . Surve i l lance of the Flow of Salmonella and Campyl obacter in A Community . PHS 2 2 3 -8 1- 7 04 1 , 1 9 8 4 . 28 . Skirrow , M . B . Camovl obacter enteritis : A " new" disease . Br . Med . J . 2 : 9 - 1 1 , 1 9 7 7 . 29 . Spika , J . s . , s . H . waterman , G . w . Goo , M . E . St . Loui s , R . E . Pacer , et al . Chl oramphenicol Â resistant salmonella newport traced throuqh hamburqer to party farms : A maj or pers istinq source o f human salmonel l osis in Cal i fornia . N . Enql . J . Med . 3 1 6 ( 1 0 ) : 5 6 5-57 0 , 1 9 8 7 . 2 9 a . Tarr , P . I . , M . A . Ne il l , D . L . Christie , D . E . Anderson . IL QQli 0 1 57 : H7 Hemorrhaqic Col it i s . ( l etter) N . Enql . J . Med . 3 18 ( 2 5 ) : 1 6 9 7 , 19 8 8 . 30 . St . Louis , M . E . , D . L . Morse , M . E . Potter , T . M . DeMe l f i , J . J . Guzewich , R . v . Tauxe , and P . A . Blake . The Emerqence of Grade A Eqqs as a Maj or Source of Salmonella Enteritidis Infections . New impl icat ions for the control of salmonel losis . J . Amer . Med . Assn . 2 59 ( 14 ) : 2 1 0 3 -2 1 0 7 , 19 8 8 .
VI I I THE ESTIMATION OF RISK RISK ASSESSMENT AND UNCERTAINTY As noted in earl ier chapters of this report , there is no direct evidence that subtherapeut ic uses of ant ib iotics in animal feed create an excess risk of disease or death in humans consuming products from treated anima l s . Thi s i s not unexpected , at the present state of knowledge it would be unreasonabl e to expect direct evidence , even i f the risk were relatively large . Attempts to col l ect such evidence are beset with serious methodol ogic d i f ficulties ; opportunities for col l ecting direct evidence on the magnitude of this risk are rare ; and when they exist , are not l ikely to produce unambiguous results . There are questions about what agency has the mandate , funding , or manpower to obtain the information needed to evaluate the ef fectiveness of any regulatory act ion regarding the feed additives under cons ideration here . The cost of the data col l ection w i l l certainly be h igh . There are unanswered questions about whether FDA now has the legal authority to demand the col l ect ion and submiss ion of the types of data that would be required to show sa fety , in its broadest sense for human health . The tool s of risk assessment are typically appl ied in situat i ons of this type where there is a need to acquire some sense of the probable s i z e of a potential publ ic health prob l em , some relevant data are ava i l able , but no means 1 exists to obtain a direct measure of risk . Because r i s k est imates produced by such mean s a re based on a s sumpt i ons and l im ited data , they shou l d be interpreted and u s ed w ith caut i on . Such est imates are best seen a s sc i e nt i f ic hypothe s e s about the pos s ib l e extent o f a prob l em . Th i s does not mean they a re " hypothe t i ca l " i n the weak sense tha t they are ba sed on specu l a t i on . Rather , they are hypothe s e s that a r e cons i stent with a l l ava i l ab l e i n forma t i on and s c i ent i f i c understand i ng , but that have not been ver i f ied by trad i t i on a l s c i ent i f i c methods . Al l the e s t imates pre s ented in th i s report shou l d be v i ewed in th i s pe r spect ive . An essent ial part of any risk assessment is the characteri z ation of the associated uncerta inties . In most cases , including the present one , risks are presented as numerical est imates ( for example , as deaths per year) or as 149