evidence (e.g., autopsy findings or operation without histologic analysis). Histopathological typing of the tumors was based on the classification of the World Health Organization (WHO).¹⁴ The reports from the pathology departments were reviewed and used to code the cancer in 1264 subjects (92.9 percent). For the remaining subjects, the coding was based on information from the Swedish Cancer Registry. This registry used a classification that was compatible with the WHO system for squamous-cell carcinomas and adenocarcinomas but did not differentiate between small-cell and large-cell carcinomas.

Two control groups representing the study base were selected from the population registers of Statistics Sweden. Each group included 1500 subjects. The first control group was frequency-matched for age (in five-year intervals) and calendar year of residence with the case group, and it originally included 775 women and 725 men. Immigrants to Sweden after January 1, 1947, were excluded, leaving 730 women and 694 men for the subsequent analyses (Table 1). The slight differences in age distribution between the case subjects and the controls resulted from the exclusion of immigrants.

The second control group was selected according to the same criteria used to select the first group, except that in addition it was frequency-matched for vital status, with use of the Swedish Cause of Death Registry. Matching for vital status was performed to reduce potential bias in obtaining information on exposure. Subjects who had died of smoking-related causes were excluded from the second control group to avoid overrepresentation of smoking.¹⁵ On the basis of evidence from Swedish studies,^16,17 the following diagnoses were regarded as related to smoking: cancer of the mouth, esophagus, liver, pancreas, larynx, uterine cervix, or bladder; ischemic heart disease; aortic aneurysm; cirrhosis of the liver; chronic bronchitis and emphysema; gastric ulcer; death from violent causes; and intoxication. The second control group originally included 683 women and 817 men, but after the exclusion of those who did not reside in Sweden on January 1, 1947, a total of 650 women and 773 men remained.

When the study subjects were selected (on December 31, 1986), 518 women (88.4 percent) and 706 men (91.2 percent) in the case group had died. In the first control group, 55 women (7.5 percent) and 68 men (9.8 percent) had died, and in the second control group, 572 women (88.0 percent) and 707 men (91.5 percent) had died.

All the study subjects or their next of kin were mailed a standardized questionnaire inquiring about the smoking habits of the subjects and their spouses and parents. The subjects’ lifetime occupational history and their residential addresses since 1947 were also investigated. Questions were asked about the type of house, the building material used, the heating system, the amount of time spent at home, and the like. In the event of an incomplete questionnaire or a failure to respond, supplementary information was obtained in telephone interviews. Those collecting the data did so without knowing whether the subject under study was a case subject or a control.

Questionnaires were returned for 1118 case subjects, as well as for 1192 and 1135 subjects in the two control groups, yielding response rates of 82.2, 83.7, and 79.8 percent, respectively. In the first control group the respondents to the questionnaire were primarily the study subjects (81.7 percent), whereas in the case group and the second control group next of kin predominated (91.8 and 90.7 percent). Among next-of-kin respondents, spouses were the most common (47.8 percent in the case group and 43.8 percent in the second control group), followed by children (39.0 and 37.9 percent in the case group and the second control group, respectively).

The assessment of each subject’s exposure to radon was based on a residential history and on radon measurements. In the compilation of the residential history, data from parish registers were supplemented with information from the questionnaires, so that a complete record of residential addresses from 1947 on was made available. The radon measurements were intended to include all dwellings in which the subject had lived during a “residential period,” defined as a period of two years or more from 1947 to three years before the end of follow-up. The year of diagnosis constituted the end of follow-up for the case subjects, whereas the frequency-matched year of selection was used for the controls.

A total of 13,392 residential periods were identified (Table 2), but for 7.5 percent the address could not be identified because the subject resided in an unknown place, abroad, in a hospital, on a ship, or the like. Information on addresses was available for 12,394 dwellings, or an average of 3.1, 2.9, and 2.8 dwellings per subject in the case group and the first and second control groups, respectively. Radon measurements could not be made in 3402 dwellings (27.4 percent), usually because the house no longer existed or was being used only as a summer house.

Radon was measured over a period of three months during the heating season—i.e., a time between October 1 and April 30. In each dwelling one detector was placed in a bedroom and another in the living room, mostly by personnel from the local board of public health. Radon was measured by solid-state alpha track detectors processed at the Swedish Radiation Protection Institute. The system includes an alpha track detector, a holder, a chemical etching process, and an automatic readout by an image system.¹⁸ For a measurement period of 90 days, the total error resulting from uncertainty in calibration, film sensitivity, readout, counting statistics, and background is 10 percent at radon concentrations of 1.6 pCi per liter (60 Bq per cubic meter), 7 percent at concentrations of 3.1 pCi per liter (115 Bq per cubic meter), and 5 percent at concentrations of 10 pCi per liter (370 Bq per cubic meter). The detectors were calibrated at the Radiation Protection Institute, which has taken part in international comparisons since the 1970s with good results.^19,20

Cumulative radon exposure since 1947 was estimated for each subject by adding the products of the measured radon level and the length of time the subject lived in each residence. Time-weighted mean radon concentrations were calculated by dividing the cumulative radon exposure by the total time spent living in residences for which radon measurements were available. In some analyses of cumulative exposure to radon, missing measurements were replaced by the median radon level for all study subjects. In other analyses, these replacements were based on information about the characteristics of the residence (the building material and type of house) obtained from the questionnaire and the characteristics of the municipality (high, medium, or low risk of radon in dwellings). Information on whether the subjects slept in a room with an open window, which may have an influence on radon exposure, was used in some analyses. Cutoff points in the analyses using time-weighted mean radon concentrations were based partly on current Swedish standards.

Smoking habits were classified according to the time-weighted mean consumption of tobacco during the subject’s lifetime. Daily consumption was expressed in cigarette equivalents, with one pack (50 g) of pipe tobacco a week corresponding to 7.1 cigarettes a day. Conversions were also made for cigarillos and cigars, which were rarely used. Subjects who stopped smoking two or more years before the end of the follow-up period were classified as ex-smokers.

Each job held by a study subject was classified in one of four categories based on earlier evidence of occupational risks of lung