Physician Staffing for the VA: Volume I (1991)

Since the study's inception, it has been clear that expert judgment would be important in the formal development of a VA physician requirements methodology.

The original statement of work noted that ''Because the available empirical data base alone is not adequate for driving the development effort or generating quantifiable estimates by purely mechanical numerical exercises, relevant informed professional judgments will be required throughout . . . and may well be an integral component of the physicians' requirements methodologies itself'' (Institute of Medicine, 1987). To implement this mandate, the committee was to appoint "advisory panels to broaden the base and range of experience and competence" brought to bear in the development of the methodology.

In response, the committee established 11 advisory panels: data and methodology (central to the analyses in chapters 4, 7, and 8); affiliations (see chapter 9); nonphysician practitioners (see chapter 10); and six specialty and two clinical program panels, to serve as sources of professional judgment in the methodology's development. The six specialty panels were medicine, surgery (including also anesthesiology), psychiatry, neurology, rehabilitation medicine (including also spinal cord injury), and other physician specialties (encompassing laboratory medicine, diagnostic radiology, nuclear medicine, and radiation oncology). The committee also appointed two multidisciplinary clinical program panels in the areas of ambulatory care and long-term care. Each panel was composed of VA as well as non-VA representatives, with the former never constituting a majority.

A central issue for the committee was determining the scope of the charge given to the specialty and clinical program panels. Two general approaches were considered:

1. In a physician requirements methodology relying primarily on the Empirically Based Physician Staffing Models (EBPSM), the panels would be asked to react to the estimated statistical models presented to them, evaluating their specification from a clinical perspective, and possibly modifying either the models themselves or their staffing recommendations.

2. In a physician requirements methodology calling for a more balanced reliance on statistically based and expert judgment-based approaches, the panels would serve as the principal source of independently derived quantitative assessments of appropriate physician staffing.

Under this second approach, the panels would not simply be critiquing and modifying statistical models, but would be rendering their own professional judgments about physician staffing levels consistent with high-quality medical care in particular clinical settings. These Full-Time-Equivalent Employee (FTEE) levels could then be compared with those emerging from the EBPSM for those same clinical settings.

Under either interpretation, the panels would seek to develop external (to the VA) physician staffing norms, which would aid in the interpretation of statistically based as well as expert judgment-based results.

The committee decided that the second, more expansive, interpretation of the panels' charge was the more appropriate.

The committee could envision a structured process in which panel members (1) are shown either a statistical model, its physician staffing implications, or both; (2) are asked to determine whether the model leads to staffing consistent with high-quality care; and (3) if not, are asked to "manipulate" the model's estimated coefficients in some fashion to generate appropriate staffing results. However, there were several concerns about proceeding this way.

First, with a single exception, the panels were constituted of expert clinicians with varying amounts of experience with formal statistical techniques; such a coefficient manipulation process would not make the best use of the collective expertise represented on the panels.

Second, given only the estimated equations (as shown in chapter 4), on what basis would panel members be able to judge whether the resulting physician staffing levels were "appropriate?" That is, can a staffing level be judged as appropriate, or not, in the absence of facility-specific information to establish a concrete backdrop—a context for evaluation?

Third, the methodological foundations for a coefficient manipulation approach have not been well established in the social science or statistical literatures. In contrast, the conceptual underpinnings and assumptions of the statistical analyses in chapter 4 are clear and well known. The expert judgment methods of decision making detailed later in this chapter, although not based on a rigorous, axiomatic approach, are nonetheless clear and unambiguous in their assumptions and implications. To enmesh the two approaches through a

coefficient manipulation process is to proceed down a methodological path whose theoretical underpinnings are not well established.¹

The committee's recommendation for how to combine, or reconcile, the empirically based and expert judgment staffing results is a choice process model termed the Reconciliation Strategy (see chapter 6).

Thus, each of the eight panels—in the course of two meetings in Washington, D.C., an extended conference call, and numerous mail and telephone communications with study staff—accomplished the following:

• Critiqued the empirically based models, offering recommendations about the choice of variables, data sets, and mathematical specification of the equations;

• Developed and evaluated external (to the VA) physician staffing norms; and

• Derived its own independent estimations of appropriate physician staffing in specific VA medical centers (VAMCs). The panels compared these results with those from the empirically based models and some external norm analysis. At that point, on the basis of the totality of evidence, the panels revised their staffing estimates accordingly.

To accomplish the latter task, the committee had to define a panel process that was methodologically sound and capable of being implemented by the eight panels in a consistent, yet flexible, way. In the health arena alone, there have been a number of recent efforts to use expert judgment processes, in scholarly analyses as well as in forums for public decision making. In the next section, the most prominent recent applications of these approaches are reviewed, and their implications for an expert judgment methodology appropriate for determining physician requirements are discussed.

In addition, details are given about what the specialty and clinical program panels accomplished in their approximately eight months of analyses in 1990. There is some discussion of their critiques of the empirically based models, which proved substantive and useful to the committee. However, this chapter focuses principally on the development of two alternative expert judgment approaches to estimating physician requirements: the Detailed Staffing Exercise (DSE) and the Staffing Algorithm Development Instrument (SADI). In addition, the process for constructing and evaluating external staffing norms is described.

In designing a process by which the six specialty and two clinical program panels would operate, the committee faced two major methodological questions. By what means and in what form would expert judgment be elicited? How would the judgments of individual panel members be combined to reach consensus positions? Before the committee's own choices are discussed, the strategies of others in this area are reviewed.

There is a growing literature on the formal use of expert opinion in health care policy and research. These applications sometimes involve the estimation of model parameters for which objective data are either missing or inappropriate. More often, expert judgment is used to reach decisions either about the advisability of particular decisions that are intermediate to a final policy outcome, or about the advisability of the outcome itself.

Although there are a number of variations on the theme, methods to elicit expert judgment in a way that leads (eventually) to consensus positions can be grouped into three broad categories: the "pure" Delphi method, group interactive methods, and modified Delphi approaches.

Panel members render judgments individually and anonymously, typically through self-administered questionnaires. The elicitation continues through several iterations. After each elicitation, the individual judgments are collected, analyzed, and fed back to all members so that each can see where he/she stands in relation to the others. The elicitations continue until, in the judgment of the analyst, either a consensus has been reached or a "point of diminishing returns is reached" (Fink et al., 1984).

The Delphi method offers several advantages. It encourages individual members to express views freely and impersonally; the opportunity is diminished for strong personalities to dominate the decision or for "group think" to lead to an artificial or premature consensus. Because the method does not require panel members to meet face to face, it can be conducted relatively efficiently and inexpensively by mail with spatially separated participants completing questionnaires on a flexible schedule.

The major disadvantage with the "pure" Delphi method is that, because panel members do not interact, there is no opportunity for each to probe the

positions of others, defend his/her own position, and thus gain a richer understanding of the problem (unless they are able to communicate informally).

Fink and colleagues (1984) cite a number of Delphi method applications in health. More recently, the Harvard-based team producing the Resource-Based Relative Value Scale (RBRVS) (Hsiao et al., 1990) has experimented successfully with several methods (including the Delphi) in developing a more efficient approach to estimating relative-value weights for surgical procedures.

Connoted here is any process in which panel members meet together, discuss information pertinent to the decision (including possibly their individual viewpoints and interpretations), and then attempt to reach a consensus.

There are several variations on this theme. Panel members may be shown background materials in advance, as with the consensus development conferences sponsored by the National Institutes of Health (Kosecoff et al., 1987). Alternatively, information for the discussion may be first revealed, or even developed, during the meeting, as in applications of the nominal group process (see Fink et al., 1984). The discussion may be wide open, so that individuals and their viewpoints are easily linked, or structured so that viewpoints are elicited anonymously and discussed without attribution.

The strengths and weaknesses of such group interactive methods are the reverse of the Delphi. The opportunity to exchange ideas can lead synergistically to conclusions in which more information has been brought to bear, in sum, than if participants had voted in isolation. But there is a risk that the outcome will be influenced by personality, meeting adjournment deadlines, and other factors that ought not to bear on the problem's resolution (although several variations of this method are designed to prevent this).

Several recent expert judgment applications have drawn selectively from both the Delphi and the group interactive approaches to evolve hybrid processes for eliciting information toward consensus development.

Most of these can be usefully characterized as estimate-talk-estimate processes (see Gustafson et al., 1973; Ludke et al., 1990). Prior to their first meeting, panel members typically are asked to render initial judgments, anonymously and independently, based on information transmitted by the analyst. These results are submitted and displayed at the first meeting. Each panel member knows his/her position relative to the group as a whole but may or may not know how other individuals, by name, have voted.

Following discussion, the group votes again; depending on the format, this poll may or may not be anonymous. Again, the results are analyzed and displayed. The process continues until the analyst determines that either a consensus has been reached or else the costs of continuing outweigh the benefits.

Such an approach draws strength from Delphi as well as group interactive methods. By first eliciting judgments anonymously, the analyst maximizes the amount of independent (judgmental) information brought to bear on the question. The opportunity to discover plausible "outlier" positions is enhanced, which reduces the chance that the subsequent consensus will be predicated on an overly restrictive conception of possible outcomes. By discussing these initial assessments in a group setting, each panel member can benefit from the views of others, thus bringing the maximum amount of (judgmental) information to bear on his/her upcoming reassessment.

On the other hand, there is the concomitant risk that personality factors, adjournment deadlines, group-think pressures, or other extraneous matters will contaminate the group interaction part of the process. The effects of these factors can be reduced by maintaining the anonymity of the panel members' positions and by such practical steps as pacing the meetings so that ample time is allowed for discussion and voting.

Studies in which a modified Delphi method has been applied include the assessments of U.S. physician requirements, by specialty, conducted initially by the Graduate Medical Education National Advisory Committee (GMENAC) (U.S. Department of Health and Human Services, 1981) and currently by the Council on Graduate Medical Education (COGME) (Buerhaus and Zuidema, 1990); the Effectiveness Initiative conducted by the Institute of Medicine to assist the Health Care Financing Administration in setting priorities for medical practice analyses (Institute of Medicine, 1989); a series of analyses to project faculty needs as well as the manpower required to care for the elderly in future decades, based at the University of California at Los Angeles and RAND (Reuben et al., 1990, 1991); a project conducted at the Iowa City VAMC examining the appropriateness of certain nonacute inpatient admissions to VA facilities across the country (Ludke et al., 1990); a portion of the RBRVS study cited earlier (Hsiao et al., 1990); and analyses conducted by RAND in recent years to determine appropriate clinical indications for performing various medical and surgical procedures (see, e.g., Park et al., 1986).

There are basically two ways of arriving at a group consensus. The participants may be formally polled and the votes aggregated in some fashion to yield a group choice, or the group may agree to hammer out a consensus position following discussions in which the relevant data and views of individuals have

been aired. Such a consensus may be explicitly declared to be unanimous, the impression may be left that it is unanimous, or dissenting statements or minority reports may be filed.

The most prominent forum utilizing the second approach is the program of consensus development conferences sponsored by the National Institutes of Health (Fink and Kosecoff, 1984). A recent survey (McGlynn et al., 1990) indicates that government-sponsored consensus development conferences in eight other industrialized nations also shy away from formal procedures for achieving agreement.

On the other hand, all other expert judgment applications cited earlier do use explicit decision rules to map individual judgments into a consensus position. Nearly all decision rules apply to one of three types of choice problems. The group must either (1) agree or disagree, or determine the extent of its agreement or disagreement, with one or more propositions; (2) develop a preference ranking for a set of items; or (3) produce quantitative estimates of variables or parameters for use in subsequent calculations, leading eventually to some research or policy conclusion.

An interesting example of (1) arises in the RAND studies on clinical indications for intervention (Park et al., 1986). In this modified Delphi approach, panelists were asked, prior to their first meeting, to rate each possible clinical indication for a given intervention (e.g., endoscopy) on a scale of 1 to 9, with 9 meaning "extremely appropriate" and 1 meaning "extremely inappropriate." When the panelists met, they were shown the resulting frequency distribution of their ratings; each panelist could see where his/her score fell relative to the group. Following discussion, they were then asked to reevaluate the indications on the same 1 to 9 scale.

Finally, whether the panel was in "agreement" or "disagreement" that a given clinical indication was appropriate was determined as follows: The high and low extreme scores were discarded, and the median of the remaining scores was computed. If these remaining scores fell within any three-point range on the nine-point scale, the panel was said to be in "agreement,'' with the median score indicating the relative degree of appropriateness/inappropriateness of the indication for the intervention in question. On the other hand, if at least one rating fell in the 1-3 range and at least one in the 7-9 range, the panel was said to be in "disagreement." Otherwise, the panel's position was said to be ''equivocal."

An interesting, though not unexpected, result in three separate evaluations was that a panel's second ratings were closer to one another than the initial ratings, whether measured by the percentage of agreement, percentage of disagreement, or average dispersion of scores.

An index of the latter is the mean absolute deviation (MAD) statistic, defined as Σ (X_i-X_med/N, where X_i is the score of the ith panel member, X_med

is the panel median score, and N is the number of panel scores used in the decision process.

A broadly similar approach to decision making was used in the VA study examining the appropriateness of acute inpatient admissions (Ludke et al., 1990).

An example of the second type of consensus choice problem is found in the IOM's Effectiveness Initiative study (Institute of Medicine, 1989), in which certain scoring rules were used to derive a priority ranking of clinical conditions for further research.

When a panel is asked to derive a best estimate of a variable or parameter that can take on many (sometimes an infinity of) possible values, how should a consensus be defined? As it turns out, this is precisely the choice problem arising in the expert judgment models developed for the present study.

In the GMENAC and COGME studies, expert panels estimated a number of parameters used in the calculation of the "adjusted need" for physicians (Buerhaus and Zuidema, 1990; U.S. Department of Health and Human Services, 1981). In GMENAC, the consensus value of any given parameter was the panel median estimate; to lend perspective, the high and low values were also reported. In COGME, a range of values are reported for each estimate of physician need or supply, and calculations involving these variables typically use the range midpoint values.

In the small group judgment study recently conducted by the RBRVS project (Hsiao et al., 1990), panels of surgeons used a magnitude estimation technique to rate the relative amount of work required to perform a number of services. At each juncture in the process of rating each service, a median score was computed. A consensus was declared whenever all scores fell within a predetermined acceptable range of the median.

In light of these studies and policy applications, the committee initially determined that the specialty and clinical program panels' own estimates of appropriate physician staffing levels would be obtained through a process with the following operating characteristics.

A modified Delphi approach would be developed in which panel members would independently estimate appropriate physician staffing levels (in the applicable specialty or program area only) at a selected set of actual VA facilities. These estimates would be tabulated by study staff and displayed anonymously to panel members when they next convened. In the course of discussions, it might become natural, or necessary, for individuals to become identified with their estimates, but this should evolve only as needed.

Following discussion of the first round of estimates, the panel would be asked to reassess physician requirements (in its specialty or program area only). These results likewise would be tabulated and displayed. In principle, the reassessments would continue until the members' physician FTEE estimates had—by some criterion—stabilized sufficiently that a panel consensus estimate could be declared.

But how should a consensus be defined?

Following each iteration of physician FTEE assessments by the panel, the median value would be computed and the high and the low values noted. By one reasonable definition, a consensus emerges when the median stabilizes. More formally, a consensus is declared on the ith iteration if the resulting median is within an acceptable range of the median obtained at the (i-1) iteration (the previous one).

A stronger definition of consensus would require that both the median and the MAD statistic, measuring here the average dispersion of physician FTEE responses around the median, not change appreciably between assessment iterations. All else equal, this more stringent definition—requiring stability in the dispersion of assessments as well as their central tendency—is preferred.

As will be seen shortly, the concepts underlying both the committee's preferred scheme for eliciting expert judgment and its preferred definition of consensus undergird the operations of the eight panels.

Given this study's developmental nature and time constraints, however, the panels' consensus assessments of appropriate physician staffing—via both the DSE and the SADI—must be regarded as approximations of what would be obtained had these expert judgment processes been able to proceed through several iterations. Again, the panels' charge in this regard was to help the committee develop methods for staffing, not to render the final numbers on VA physician requirements.

In this section the operation of the six specialty and two clinical program panels is described in terms of what turned out to be their major functional responsibilities: evaluating the EBPSM, developing and testing the DSE, developing and testing the SADI, and evaluating external (non-VA) norms to guide physician staffing decisions.

The primary focus here will be on the DSE and SADI because they are new vehicles for deriving expert judgment estimates of appropriate physician staffing; as such, they played central roles in most of the panels' recommendations for how the VA ought to determine physician requirements.

Although the planning for panel operations began early in the study and their interactions with the committee and the staff continued through the first six

months of 1991, the bulk of the activities described below occurred during the first 10 months of 1990. For expository purposes, it is useful to divide this period roughly into three phases: preparation for and conduct of the first panel meetings (January through April); preparation for and conduct of the second panel meetings (May through mid-August); and postmeeting activities (mid-August through October), culminating in a panel chairmen's session at the November 1-2, 1990, meeting of the committee.

Before the panels' accomplishments are discussed, the procedures for appointing panel members are reviewed briefly.

The committee intended that the membership of each panel reflect a broad spectrum of clinical knowledge, professional judgment, and special technical expertise. Collectively, the physicians on each panel were selected to bring perspectives spanning a variety of clinical practice settings.

It was understood from the beginning that the study would focus on the major specialty and program areas prominent in the VA; hence, the committee was constituted so as to have representation in these areas. It was natural that the chairs of the six specialty and two clinical program panels be drawn directly from the committee membership.

The study's workplan called for each panel to consist of VA as well as non-VA members, with the latter constituting a voting majority in each case.

In response, the committee asked the Department of Veterans Affairs to nominate VA staff candidates for panel membership. The VA liaison committee proposed candidates for each panel, and a list of nominees was subsequently submitted to the IOM by the VA chief medical director.

Non-VA panel nominees were initially solicited from members of the study committee. Additional nominees were drawn from the IOM membership, in consultation with the director of the Division of Health Care Services and the IOM executive office.

After all nominations were received, a tentative panel roster (of non-VA and VA candidates) was submitted to each panel chairman for review. Each chair could propose additional nominees. The final selection of VA and non-VA members was made by the panel chairman in consultation with the chairman of the committee. (A complete set of panel rosters is contained in Appendix A of this report.)

The specialty and clinical program panels provided important critical advice to the data and methodology panel and the committee about several aspects of the EBPSM:

At its first and second meetings and during the postmeeting period, each panel was shown various specifications of empirically based models pertinent to its specialty domain or program area. Each was asked to address several questions:

• Is workload defined appropriately?

• Are the physician FTEE variables properly constituted?

• Do the variables included in the equations make clinical and organizational sense?

• Did the variables perform as expected statistically?

• For coefficient estimates that are not statistically significant, or that are significant but with the "wrong" algebraic sign (indicating "perverse causality"), what factors might be at work?

• Are there variables currently omitted from the equations that should be tested on clinical or organizational grounds?

During the first and second panel meetings, the panels' empirically based model critique focused entirely on the production function (PF) variant. The inverse production functions (IPFs) did not begin emerging until the postmeeting period and were then evaluated by the panels at two junctures: first, via mail communications with study staff during late August; and, second, during the conference calls with staff in late October.

In the course of these meetings, written communications, and phone calls, panel members contributed numerous suggestions on improving the empirically based models (including the sentiment, expressed on occasion, that the models be discarded entirely in favor of an expert judgment approach). There was not a panel whose empirically based models were not significantly modified as a result of these give-and-take discussions.

During both its first and its second meetings, each panel examined the model-derived physician FTEE level and the corresponding level for three selected VAMCs (four for psychiatry in its second meeting), as recorded in their cost distribution reports (CDR). The panel could compare these estimates with those it derived for the same facilities using the DSE. During the postmeeting period, panel members were also shown the physician requirement estimates for these facilities as derived from both the IPF and the SADI.

Given this array of alternative FTEE estimates, each panel member was asked to state (during the postmeeting period via mail survey) what he/she regarded as the most appropriate physician FTEE level for each facility. Among other questions posed was whether the VA should adopt a physician requirements methodology whose centerpiece is an empirically based approach (either the PF or the IPF).

The panels' responses to these questions, and many more, are summarized in the appendix to chapter 6. A complete description of each panel's activities and policy recommendations is found in Volume II, Supplementary Papers.

In developing an approach for determining physician requirements solely on the basis of the consensus judgment of designated experts, the committee had to resolve several issues:

• How should the experts be selected?

• How should a consensus judgment about physician staffing be defined?

• How does one characterize an "ideal" mechanism for obtaining expert judgment on physician staffing?

• Can a practical mechanism be developed for eliciting expert judgment so that resulting FTEE estimates approximate the ideal? Can these estimates be obtained with sufficient specificity that physician requirements can be validly computed for any specialty or program area at any VAMC?

Suppose an expert panel is charged with determining physician requirements for a given specialty or program area at some VAMC. An ideal expert judgment mechanism is one that yields the same staff physician FTEE levels that would be derived if the panel had made the assessment with "complete information" about

the volume and the severity of patient workload, the number and the type of residents and nonphysician personnel, other facility-specific data, and the relationship between staffing patterns and indicators of the quality of care.

In preparation for the first round of specialty and clinical program panel meetings, study staff developed first-generation versions of the physician staffing instruments that would evolve into the DSEs. For each panel, three distinct (though generically similar) instruments were constructed for each of three VAMCs; the VAMCs were selected to reflect diversity in geographic location and affiliation status (as indexed by bed size, residents, and scope of services offered).

Each panel analyzed the same three VAMCs. Prior to the first panel meeting, members were mailed the instrument for one of the facilities and asked to complete it as a "homework exercise." The identity of the facility was not revealed. At the first meeting, staff presented a summary of these homework results, they were discussed, and the panel completed the other two exercises. For one of the latter, the panel was divided into small groups of two to three members each; for the other, the panel worked together as a single group. In neither case was the identity of the VAMC revealed until the exercises were completed.

Because these first-generation instruments were indeed transitional documents soon to be revised (into what later became the DSE), they are not discussed in depth here. Similarly, because the panels' staffing analyses during the first meeting were entirely exploratory, the resulting physician FTEE estimates are sufficiently experimental that policy inferences are not meaningful.

At the May 1-2, 1990, meeting of the committee, each panel presented a brief progress report. After discussions, the committee concluded two things: First, determining physician requirements by an expert judgment process was feasible, from a cognitive as well as a group dynamics standpoint. Second, the initial instruments needed revision, aimed primarily at providing enough context-specific information that panel members could assess, with confidence, physician requirements for a given ward, clinic, or program at a given VAMC.

Specifically, at the May 1-2 meeting, the committee directed each panel to work with staff to develop and test a revised instrument, subsequently termed the DSE. Each DSE consists of an A and a B section. Section A provides a ward-by-ward, clinic-by-clinic description of the patient care environment at an actual

VAMC. For each patient care area (PCA, as defined in chapter 4), information is provided on the volume and diagnosis-related group (DRG) mix of workload, number of residents by specialty and postgraduate year (PGY), number of nonphysician practitioners by type, general information about the adequacy of nursing and support staff, and other contextual details.

For each ward, clinic, and procedure, the expert is asked to assess the amount of physician time—in hours—required per day, per visit, or per unit, respectively, to produce good-quality care.

The B section contains questions on the amount of physician time required for night and weekend coverage on the PCAs, educational activities not occurring on the PCAs, research, administration, other facility-related activities, and leaves of absence.

For each panel member, time estimates for all patient care and non-patient-care activities are summed and converted to FTEE—assuming one FTEE translates into a 40-hour/week commitment.

The panel consensus estimate is defined as the median of the members' estimates.

The format for the second panel meetings was as follows: Staffing instruments for two VAMCs, hereafter identified as I and II, were mailed to panel members in advance. Neither facility was identified by name at this point. As before, each panel assessed physician requirements for its specialty or program area only. With some exceptions, panel members did complete and return both instruments to staff prior to the meeting. (Also prior to the meeting, each member received by mail a briefing book, containing the meeting's agenda, background reading, and staffing analyses relevant to the upcoming discussion.)

The members' (initial) physician staffing assessments were tabulated, checked for arithmetic errors, and presented at the panel meeting in a way that kept the members' assessments anonymous. (During subsequent group discussions, however, members typically revealed their assessments.) For VAMCs I and II, in turn, the panel was shown each member's physician activity time assessments for the component parts of sections A and B and for total physician FTEE required at the facility. Also presented were some summary statistics: the panel's high, low, mean, and median estimates of total physician FTEE.

Following discussion of these results, the panels reassessed physician requirements for VAMCs I and II, working from copies of the staffing exercises they originally submitted. For seven of the eight panels, members reassessed independently; the results were tabulated with mean and median computed, then discussed. The surgery panel determined that it could derive consensus time estimates most efficiently through a group interactive process, in which the panel as a whole discussed each FTEE component of the DSE, arriving in each instance at an estimate agreeable to the group. Such an approach leads directly to consensus estimates for total surgeon and anesthesiologist FTEE required at

VAMCs I and II; the panel's high, low, mean, and median converge to a single FTEE estimate.

At this point, each panel was shown the names of VAMCs I and II; their actual, CDR-recorded physician FTEE levels for FY 1989 for the specialties or programs within the panel's domain; and the corresponding calculation of physician requirements derived from the PF variant of the empirically based model.²

Next, the panel was asked to assess physician requirements at another VAMC, hereafter known as VAMC III, using a DSE prepared for the task. (The psychiatry panel also analyzed a fourth facility, VAMC IV.) For all panels except surgery, the assessments were again completed independently, the results were tabulated, and the mean and median were found. The surgery panel again elected to reach consensus directly through structured group discussion.

Following these assessments, staff revealed the identity of VAMC III (and, for psychiatry, VAMC IV), its CDR-recorded physician FTEE level in FY 1989, and the corresponding FTEE calculation derived from the PF model. As time permitted, there was general discussion about determining physician requirements through the DSE approach.

The best way to gain a clear understanding of how the DSE works is to examine a completed instrument in some depth. In Figure 5.1, the medicine panel's DSE is presented as constructed expressly for VAMC II.

The physician time estimates shown to illustrate the process are the initial "homework" assessments of a medicine panel member who was particularly conscientious about documenting his assumptions and reasoning. This panel member's calculations and accompanying commentary serve to make Figure 5.1 reasonably self-explanatory.³

Tables 5.1 and 5.2 indicate how physician requirements at a VAMC can be determined via application of the DSE. Table 5.1 summarizes each panel's initial (premeeting) assessments of FTEE requirements for VAMC II; Table 5.2 summarizes each panel's reassessed values following discussion of the initial results at its second meeting.

In both cases, the high and low estimates made by panel members, the panel median, and the MAD statistic indexing the relative dispersion of individual estimates about the median are presented.⁴ For some panels (e.g., ambulatory), considerable dispersion may remain.

In surveys completed during the postmeeting period, a majority of the members of all eight panels concluded that the DSE offers a technically feasible and methodologically acceptable expert judgment approach for deriving physician requirements. (See the appendix to chapter 6 for the panels' concluding statements and Volume II, Supplementary Papers, for each panel's full report to the committee.)

But the committee notes, as did some panel members, that the DSE is also cumbersome, labor intensive, and not well suited in its current form (as depicted in Figure 5.1) for probing what physician staffing ought to be under alternative assumptions about workload, nonphysician resources, and other factors.

To remedy the latter is straightforward but would likely require ever larger, more complex DSE instruments, with corresponding increases in respondent burden. Moreover, for a systemwide application of the methodology, each VAMC would require its own set of DSEs, which would have to be individually evaluated by panels convened each time a new assessment was required. Building a streamlined, efficient physician requirements methodology around the DSE approach appears to be problematic.

In response, work began in August 1990 on an alternative expert judgment approach within each panel. The committee wanted to build upon, formalize, and strengthen some of the rules of thumb for staffing evident during the second panel meetings. Ideally, this new approach would retain much of the DSE's specificity, while being more streamlined and less labor intensive in application.

By late August, first-generation versions of the SADI had been developed for experimental application by each of the eight panels. By early September, the members of each panel, working separately and independently, had completed and returned their SADIs by mail to study staff. The results were processed, and first-generation SADI-based estimates of physician requirements were computed for VAMCs I, II, III, and (for psychiatry) IV.

An overview of the SADI approach and how it was applied by the panels follows. For illustration, the medicine panel's SADI is presented, and that panel's physician activity and time estimates are summarized. Following that, the SADI approach is applied to derive physician requirements in medicine at a given facility (VAMC I). Next are presented the FTEE levels that emerged when each panel's SADI estimates were applied, in turn, to determine physician requirements at VAMC II. Finally, the committee evaluates the SADI approach and offers recommendations for its further development.

Several specific steps were involved in applying this new expert judgment approach within each panel. As will be seen, the SADI (like the DSE) has two sections, A and B, focusing on patient care and non-patient-care activities, respectively.

In section A, panel members were asked to estimate the amount of physician time required to perform each of an array of functions and tasks in a way consistent with achieving good-quality care. Patient workload categories for which physician time estimates (typically expressed in hours) were sought include inpatient admission workup, routine daily care on the wards of that specialty's or program's ''dominant'' PCA (if applicable), consultations on all other PCAs, certain diagnostic and therapeutic procedures, and outpatient visits. In each category, physician time can be estimated as a function of the availability of residents by type, nonphysician practitioners by type, and other contextual factors.

In section B, panelists were asked to determine the total amount of physician time (again, for physicians in the panel's domain) that ought to be devoted to the following non-patient-care activities: research, education of residents in the classroom, continuing education, administration, other hospital-related activities,

and leaves of absence. The format in section B of the SADI is virtually identical to that used in the final version of the DSE.

In deriving group consensus SADI estimates from individual member estimates, each physician task or function in section A was considered in turn: the individual estimates were arrayed, and the median was designated as the panel's consensus estimate. Likewise, for each FTEE component of section B, the panel median was declared the consensus estimate.

To determine physician requirements at a given VAMC, the panel median for each type of patient care activity was applied, in turn, to the volume of such activities associated with the facility's projected workload, and the results were summed across activities to derive total physician hours for patient care. Physician times for all non-patient-care activities were assessed separately using the relevant median estimates. The sum of physician hours estimated from sections A and B was then converted into FTEE using the 40-hour/week equivalence assumption.

As with the DSE, the best way to understand the SADI is to examine a completed instrument, then study how it can be applied to determine physician requirements at some VAMC.

In Figure 5.2, the medicine panel's SADI is presented in its entirety, indicating for each function or task that panel's high, low, mean, and median estimates of the amount of physician time required for good-quality care. The instrument presented here is a slightly compact version of the one completed by seven of the panel's eight members; it is intended to be self-explanatory.⁵

Although each panel's SADI is tailored specifically to the main activities associated with that specialty or program area, all SADIs are basically similar both in structure and the logic of application; to understand the medicine instrument and how it is applied is to understand the SADI approach.

The application of this medicine SADI to determine physician requirements at VAMC I is summarized in Figure 5.3, which is intended to be a relatively self-contained walk-through of how the SADI works. However, several general points deserve emphasis.

For each physician activity (e.g., admission workup), the panel consensus estimate of the time required per unit (e.g., per workup) is multiplied by the projected number of units of the activity per day. One exception is the time required for routine daily patient care on the wards, which comes packaged as a total that varies with the average daily census (ADC) and other contextual factors.

Total physician hours required per day is the sum of hours required for all patient care and non-patient-care activities. A final step is converting hours into an FTEE equivalent.

In the SADI (and the DSE, too), the whole is defined as the sum of the parts; hence, the parts must successfully encompass all physician activities at (or associated with) the VAMC.

Data Required from the VAMC. To apply the SADI in its current form to determine physician requirements at a VAMC, the facility itself must generate certain workload and other data; the VA's central information systems do not generally supply information at the level of detail required by the SADI (or the DSE). These information requirements can be inferred from Figure 5.3, but are summarized here as well.

• Patient Care Activities—inpatient:

Admissions per day in the specialty's dominant PCA (if applicable), with and without resident;

For each ward in the dominant PCA, the ADC, the average length of stay (LOS), and the number and type of residents as a function of ADC;

For each special care unit where the specialty is a major participant (e.g., ICU/CCU for medicine), the ADC, the LOS, and number and type of residents as a function of ADC;

For all other wards, including intermediate care and nursing home, the number of initial as well as followup consultations per day, with and without resident;

The number of special procedures (e.g., cardiac catheterizations in medicine) performed per day, with and without resident; for both surgery and anesthesiology, the number of operations per day distinguished, as the surgery panel has recommended, by level of complexity.

• Patient Care Activities—ambulatory:

The number of patient visits per day by ambulatory PCA, with and without resident, and with and without physician assistant or nurse practitioner.

• Non-Patient-Care Activities:

Sufficient information about total research funding at the VAMC that it can be classified as either a high-, medium-, or low-volume research facility—the overall proxy for affiliation status used in section B of all current SADIs (see Figure 5.3).

As it turns out, all of this information (and then some) is required for constructing the DSEs; thus, the facility-specific data for implementing the SADIs here were extracted directly from the existing DSEs.

Interpolation or Extrapolation. The projected ADC for a given ward at a VAMC may not match exactly any of the ADC levels for which time estimates are available in the SADI, as evident in the Routine Daily Patient Care portion of Figure 5.2. For example, if the projected ADC is 35 and the highest ADC shown in the SADI is 30, an extrapolation is required to estimate physician time for 35. Similarly, if the projected ADC is 22 and the nearest ADC levels included are 20 and 25, physician time for 22 must be interpolated.

In both cases, the simplest approach would assume a linear relationship between ADC and physician time. This may or may not be warranted. For example, suppose the estimated median times for the ADCs of 20, 25, and 30 are 4, 5, and 5.5 hours, respectively. For an ADC of 22, an interpolated estimate of 4.4 hours—i.e., 4 + [(22-20)/(25-20)]—seems reasonable. But given the nonlinear way physician time responds as ADC goes from 20 to 25 to 30, it seems unreasonable to calculate the time for an ADC of 35 as (35/30)5.5 = 6.4 hours; rather, something less than this is more plausible.

If SADI estimates are directly available for a sufficiently dense set of ADC levels, the issue becomes moot. Not surprising, there is a trade-off between the level of detail built into the SADI—and thus the resources required to produce and maintain the SADI—and the likelihood of having to estimate physician times from those explicitly available in the instrument.

Estimation of Physician Times for Activities Not Considered in the SADI. For example, at VAMC I there is a bone marrow transplant unit (BMTU), but the current medicine SADI includes no such activity category. Therefore, physician time estimates for the ICU/CCU unit were applied to the ADC levels projected for the BMTU. As indicated in Figure 5.3, this type of approximation was required in several instances.

Again, if the SADI is constructed in great enough detail—if all relevant activity categories are included—such approximations are not needed.

Uncertainty in SADI Activity Time Estimates. All time estimates from Figure 5.2 are treated in Figure 5.3 as if they are deterministic (nonprobabilistic) values. In reality, the amount of physician time required to perform a given activity will fluctuate for a host of reasons: differences in patient mix and acuity levels, differences in the quality and availability of nursing and support personnel, variations in patient scheduling, and variations in physician skills and pace of work. Moreover, even if these attributes were fully known and specified in a given instance, there would remain a residual amount of uncertainty about the amount of physician time required.

One response is to conduct a form of sensitivity analysis in which one or more activity performance times are systematically varied across their ranges, from high to low, and the net effect on total physician time is noted. It would not be difficult (only laborious) to extend Figure 5.3 to accommodate a sequence of such analyses.

A second approach is to acknowledge the uncertainty formally through a rigorous statistical analysis.

Assuming (as has been done throughout) that the only data sources for activity time estimates are the panel members' experience-based judgments, this new analysis would require that:

• Each panel member express each activity time estimate not as a simple point estimate, but as a (subjectively) estimated probability distribution;

• The members' individual distributions be combined to derive a corresponding panel consensus probability distribution for the time required to perform the activity;

• This consensus distribution be combined with the projected workload rate associated with the activity at a given VAMC to produce a probability distribution for the physician time required for the activity at the VAMC; and

• These activity time distributions be combined across all activities at the VAMC to derive a final distribution for the total amount of physician time, and hence FTEE, required (as always, in a given specialty or program area).

In Volume II, Supplementary Papers, this process is illustrated using the probability distribution assessments of two members of the Medicine Panel to derive the probability distribution of total physician FTEE requirements in Medicine at VAMC I. Also discussed in the Supplementary Papers is how to combine (through Bayesian statistical analysis) the panel members' subjective probability distributions with other, "objective" data on activity times from time-motion studies (or other sources) to produce new distributions reflecting both types of information. Similarly, physician time estimates from the DSE can be treated in a probabilistic fashion.

To illustrate the SADI process, the medicine instrument has been used exclusively; but the other seven panels also produced SADIs, tailored to the specialties or VA program areas within their domains. The physician FTEE levels obtained from applying these SADIs to the workload data from VAMC II in FY 1989 are summarized in Table 5.3. With few exceptions, these SADI estimates are roughly comparable to the median DSE reassessments reported in Table 5.2. There appears to be no tendency for one approach to be consistently above or below the other, across specialties.

Considering the factors noted thus far, the committee concurs with its specialty and clinical program panels that any expert judgment component in the VA physician requirements methodology should be built around application of the SADI approach, across specialties, programs, and facilities. The SADI is capable of capturing almost as much clinical detail as the DSE and is better suited for systemwide application. Given the analyses presented in the appendix to chapter 6 and the full panel reports in Volume II, Supplementary Papers, the committee regards these initial SADIs as first-generation instruments, requiring additional exploration and development.

Correspondingly, physician FTEE levels emerging at the moment should be regarded as first-generation estimates, which may change as the SADI evolves. In particular, because the SADI approach emerged late in the study, it was not feasible to use a modified Delphi process to derive physician activity time estimates. Instead, these SADI estimates are based on staffing judgments elicited through one mail survey of all panel members; in a sense, they can be viewed as the results from the initial iteration of a modified Delphi process.

To improve this first-generation model, the committee recommends the following:

The VA should proceed immediately to apply these SADIs experimentally to all VAMCs, or at least a large representative sample. For the four VAMCs analyzed in this study, staff members were able to obtain the required facility-specific workload and related data by phone and mail in a matter of days. The facilities agreed to participate voluntarily, without a formal directive from the VA Central Office.

Following an evaluation of these applications, each SADI would be considered for revision. The focus would be on

• Appropriate designation of activity time categories, with special attention to new programs and services (e.g., hospital-based home care);

• Appropriate specification of the type and range of workload for each activity category, with special attention to whether case acuity is sufficiently differentiated;

• Adequate delineation of the most salient factors influencing physician productivity, such as residents (by specialty and PGY), nonphysician practitioners (by type), nursing and support staff, and certain items of capital equipment; and

• Distinguishing among physicians by type: full time or part time; staff or contract or non-VA consulting.

Following an evaluation of these revised SADIs, they should be considered for formal application in the methodology, in ways described in chapter 6. In each case, the challenge is to construct a SADI with enough detail to capture significant distinctions, while omitting factors that have little influence on physician time allocations.

One other major issue that the committee asked the specialty and clinical program panels to investigate was whether there exist non-VA physician staffing standards or patterns that could be usefully applied to help determine appropriate VA physician staffing.

Working with study staff, each panel developed over the course of its deliberations external norm information of two types:

• Explicit physician staffing standards, primarily from a few large organizations which, like the VA, plan and deliver health care through a centralized decision-making process. The prime examples in this study were the U.S. Department of Defense (DoD, 1989) and the Indian Health Service (IHS, 1988).

• Implicit physician staffing standards, as inferred from existing secondary data, that describe the ongoing relationship between workload and physician staffing in any health care organization selected for comparison with the VA, such as an HMO or a large public hospital system.

In addition, some panels were able to bring to bear existing (though sometimes dated) physician staffing guidelines developed within the VA. Where

they existed, these internal norms provided a potentially useful basis for comparison.

All committee and panel members were asked to recommend health care organizations that might provide data relevant to the construction of norms. Several criteria were relevant: the perceived quality of care provided by the organization; the comparability of the organization's patient population and patterns of care, relative to the VA; and the quality and accessibility of its data on patient workload and physician FTEE. Guided by these factors, the study staff also pursued possible data sources.

The major organizations contacted are listed in Table 5.4; in addition, data were obtained from a number of individual hospitals, long-term care units, and clinics. The particular set of organizations analyzed by each panel is discussed in its full report to the committee, found in Volume II, Supplemental Papers.

Given the nature of (a) explicit staffing guidelines at organizations such as DoD or IHS and (b) secondary data on staffing available from other types of providers, the process for calculating VA physician requirements on the basis of external norms was technically straightforward and basically similar for all panels.

Specifically, an organization's staffing standard for inpatient or long-term care was generally defined in terms of its ADC per physician FTEE (ADC/phy)—either as posited by the organization or as observed there. An organization's standard for ambulatory care was based on either its posited or its observed ratio of patient visits per physician FTEE per year (visits/phy/yr).

As one example, consider the DoD, where explicit standards for workload per physician per year are specified separately by specialty category and by type of hospital (teaching versus nonteaching). For a facility with an ADC of 100 and 60,000 ambulatory visits per year, the physician staffing standards in medicine for inpatient and ambulatory care are, respectively, 10.3 ADC/phy and 5,808 visits/phy/yr (DoD, 1989). Applying these staffing ratios to the assumed workload data, it is calculated that

On the basis of data supplied by the New York Health and Hospitals Corporation (1989), study staff derived the following implicit physician staffing standards for Medicine: 7.0 ADC/phy for inpatient and 4,270 visits/phy/yr for ambulatory care. When these ratios are applied to the fictitious workload data above, the physician FTEE required for inpatient and ambulatory care are 14.3 and 14.1, respectively, for a total FTEE requirement of 28.4.

When the DoD and New York Health and Hospitals Corporation ratios were in fact applied to the FY 1989 workload data (suitably aggregated) at VAMC II, the implied physician staffing in medicine was 53.0 and 56.0 FTEE, respectively. (For a description of these and all other external norm calculations, see the panels' reports in Volume II, Supplementary Papers.) These estimates can then be compared with physician requirements in medicine at VAMC II as assessed through the SADI (54.0, from Table 5.3) and the DSE (49.9, from Table 5.2), as well as with the medicine staffing actually there (45.7, as recorded in its CDR).

The committee notes, however, that external norms of the type illustrated here are easy to use precisely because they involve simple (unconditional) staffing ratios. There is no control for differences in case mix, case acuity, the precise definition of an FTEE, and other factors distinguishing the source of the norm from its site of application (the VAMC). In the next chapter, the committee discusses how these factors combine to limit the usefulness of external norms at present; proposals for further development are recommended.

Buerhaus, P.I. and Zuidema, G.D. 1990. On the supply of physicians. Archives of Surgery, 125:1425-1429.

DoD (U.S. Department of Defense). 1989. Joint Health Manpower Standards. DoD #6025.12-STD, Washington, D.C., November.

Fink, A., Kosecoff, J., Chassin, M., et al. 1984. Consensus methods: Characteristics and guidelines for use. American Journal of Public Health 74:979-983.

Gustafson, D., Shukla, R., Delbecq, A., et al. 1973. A comparative study of differences in subjective likelihood estimates made by individuals, interacting groups, Delphi groups and nominal groups. Organizational Behavior and Human Performance 9:280-291.

Hsiao, W.C., Braun, P., Becker, E.R., et al. 1990. A National Study of Resource-Based Relative Value Scales for Physician Services: Phase II. Final Report. Department of Health Policy and Management, Harvard School of Public Health, Harvard University.

IHS (Indian Health Service, U.S. Public Health Service, Department of Health and Human Services). 1988. Allocation of Resources in the Indian Health Service. A handbook on the Resource Allocation Methodology (RAM), April.

Institute of Medicine. 1987. Study Workplan (Statement of Work) for a Study to Develop Methods Useful to the Veterans Administration in Estimating its Physician Needs. Washington, D.C. Unpublished.

Institute of Medicine. 1989. Report of a Study. Effectiveness Initiative: Setting Priorities for Clinical Conditions. Washington, D.C.: National Academy Press.

Kosecoff, J., Kanouse, D.E., Rogers, W.H., et al. 1987. Effects of the National Institutes of Health Consensus Development Program on Physician Practice. Journal of the American Medical Association 258:2706-2713.

Lomas, J., Anderson, G., Enkin, M., et al. 1988. The role of evidence in the consensus process. Journal of the American Medical Association 259:3001-3005.

Ludke, R.L., Wakefield, D.S., and Booth, B.M. 1990. Final Report. Pilot Study of Nonacute Utilization of VAMC Inpatient Services. Iowa HSR&D Field Program, Department of Veterans Affairs Medical Center.

McGlynn, E.A., Kosecoff, J., and Brook, R.H. 1990. Format and conduct of consensus development conferences. International Journal of Technology Assessment in Health Care 6:450-469.

New York City Health and Hospitals Corporation. 1989. Report of the New York City Health and Hospitals Corporation. In: The Mayor's Management Report to the City of New York. February. Unpublished.

Park, R.E., Fink, A., Brook, R.H., et al. 1986. Physician ratings of appropriate indications for six medical and surgical procedures. American Journal of Public Health 76:766-772.

Reuben, D.B., Vivell, S., Zwanziger, J., Bradley, T., and Beck, J.C. 1990. Study of the Adequacy of the Supply of Geriatrics Faculty at All Levels of Medicine Education. UCLA/RAND Contract No. 240 BHPr-1(9). Health Resources and Services Administration, Washington, D.C.

Reuben, D.B., Bradley, T.B., Zwanziger, J., Fink, A., Vivell, S., Hirsch, S.H., and Beck, J.C. 1991. How Many Physicians Will Be Needed to Provide Medical Care for Older Persons? Physician Manpower Needs for the Twenty-First Century. Multicampus Division of Geriatric Medicine and Gerontology, UCLA School of Medicine. Unpublished.

U.S. Department of Health and Human Services. 1981. Summary Report of the Graduate Medical Education National Advisory Committee to the Secretary, Department of Health and Human Services, September 30, 1980. Vol. I. GMENAC Summary Report. DHHS Publ. No. (HRA) 81-651. Washington, D.C.: Government Printing Office.

Figure 5.1: Detailed Staffing Exercise (DSE) for the Medicine Panel and One Member's Response¹

In the previous meeting of the medicine panel, we asked you to estimate physician staff requirements in your specialty, for a real VA Medical Center (VAMC), in a number consistent with good quality of care. This new staffing exercise repeats the process, but we have provided a more specific description of the facility including the type of patients, number of admissions, and length of stay. We have also provided more details on the number and types of residents (including level of experience and specialty) available, and the numbers of any nonphysician practitioners (NPP) that may be present. While obtaining your numerical estimates for this facility, we also will be probing the thought processes used in determining physician staffing.

This highly affiliated VAMC participates in a moderate amount of research and is large with total operating beds of 978 and an average daily census of 772.

Your task in section A is to calculate the physician hours required from the Medicine Service for each Patient Care Area (PCA) for an average weekday.

Do include in section A:

• Physician time spent on direct patient care

• Physician time spent on patient-care-related activities such as:

  • chart documentation

  • related telephone communication

  • patient and family teaching and counseling

• time spent interactively with residents in patient care and/or teaching on the PCA

Do NOT include in section A:

• Night call and weekend coverage

• Physician time spent in non-patient-care-related activities such as:

  • research off the PCA

  • educational activities that are not related to direct patient care (such as teaching residents, or delivering lectures off the PCA)

  • quality assurance

  • mortality and morbidity meetings or studies

  • administrative activities

  • any other function that is not directly related to the care of the patients on the PCA.

In the first part of section A, you must estimate the physician hours spent by medicine service physicians for an average weekday. Next, you must estimate time spent by physicians from the Medicine Service on other PCAs in the hospital, usually as a consultant. You may assume that the Medicine Service receives adequate consultative support from all other services at this VAMC.

You may assume that the level of nursing staff and support staff is adequate for this VAMC.

Appendix 1 [not included here] provides you with a list of the top DRGs and frequency of their occurrence for each individual PCA at this hospital, so that you may get a sense of the facility's case mix.

In section B, you will assess the amount of physician time that is not addressed in section A, such as non-patient-care-related activities off the PCA, night and weekend coverage, and administrative functions.

Figure 5.2: The Staffing Algorithm Development Instrument (SADI) for Medicine: The Complete Instrument with Statistical Summary of Panel's Assessments

The purpose of the SADI is to gather the data needed to construct, test, formalize, and enhance the algorithms and rules of thumb for staffing that emerged from prior meetings of each specialty program panel. The ultimate intention is to develop algorithms that could be applied to estimate staffing requirements at VA medical centers (VAMCs), presumably duplicating the results that specialty panelists themselves would have derived.

Section A of the SADI requests time estimates in some cases by workload unit. In other cases, it requests time estimates by major job elements (tasks). These elements had previously been indicated by some panel members as accounting for the bulk of the work of VA internists. For the latter cases, we seek your estimates of how physician requirements vary with respect to such variables as the volume of patients and the availability of residents and nonphysician practitioners. By systematically varying the levels of workload and nonphysician personnel, we hope to infer from your numerical responses the implicit formulas you used to relate physician time to these variables as well as the nature of the relationship between workload and staffing, e.g., linear or nonlinear.

Section B requests your response to a series of questions for the time spent in activities other than direct patient care.

Instructions: Section A: For each cell of each table, please estimate the number of physician hours required from the Medicine Service to deliver good-quality care under the specified circumstances. Section B is self-explanatory.

Please fill in the average time in hours required by a staff physician in your service to accomplish an admission work-up, either with or without a resident in your service.

For each workload factor and alternative average daily census (ADC) level below, please fill in the average number of physician hours required from the Medicine Service. Keep in mind that the daily rounds do not include new admission work-ups, since they are covered in Chart 1.

For ADC of 15 or less, assume one PGY 1 resident and one PGY 2 or 3 resident. For ADC greater than 15, assume two PGY 1 residents and two PGY 2 or 3 residents.

Fill in the average time in hours required by a staff physician in your service for each initial (new) consult on another service, noting the presence or absence of a resident in your own service. (When the resident is present, assume that he/she is performing the consult under the supervision of an attending physician.)

Fill in the average time in hours required by a staff physician in your service for each follow-up consultation visit on another service, noting the presence or absence of a resident from your service.

Please fill in the average time in hours required by a staff physician in your service for each of the special procedures listed on the left, noting the presence or absence of a resident.

Please fill in the average time in hours required by a staff physician in your service for the average ambulatory care clinic visit by a typical patient to one of your specialty program clinics, noting the presence or absence of residents and nonphysician practitioners (e.g., a physician assistant [PA] or a nurse practitioner [NP]), and whether the visit is by a new or returning patient.

Part 1. The activities listed below generally do not occur every day, but may be time-consuming when looked at over a longer period, such as a week or month. List the time in hours that you would add to each physician's average workday to allow for the types of work other than direct patient care listed below.