Duncan C. Thomas

Predicting Survival in Adults with End-Stage Renal Disease: An Age Equivalence Index

To quantify prognosis in patients with end-stage renal disease, we evaluated pretreatment clinical state and ascertained the outcome of all 220 patients who began therapy at two hospitals from 1970 to 1975. Each of three pretreatment characteristics made a statistically significant independent contribution to the relative risk of death: age (relative risk for 10-year increments = 2.2, p less than 0.001); duration of diabetes (relative risk for 10-year increments = 2.2, p less than 0.001); and left-sided heart failure (relative risk = 2.0, p less than 0.001). We combined the effects of these factors in an age-equivalence index that showed a strong gradient in survival rates from lower to higher values; the 5-year survival rate differed between 92% in patients with a score of 30 or less and 6% in patients with a score over 70. This index, which is simple to use, should prove helpful in patient care and can improve the scientific validity of therapeutic comparisons in patients with end-stage renal disease by identifying and adjusting for the selection biases that occur in the allocation of different treatments.

The relative efficiencies of matched and independent sample designs for case-control studies

We have studied the asymptotic and small sample efficiencies of dependent (pair-matched or stratified) and independent samples as design techniques for case-control studies, and of matched, stratified, covariance-adjusted, and crude comparisons as methods of analysis. The asymptotic efficiencies of dependent sample designs relative to independent sample designs with adjustment were found to vary with the strengths of the relationships of disease with exposure and potential confounder: as the relationship with exposure increases, dependent samples lose efficiency; as the relationship with confounder increases, dependent samples gain efficiency. The relative efficiency also depends in a complicated manner on such other factors as the distribution of exposure and the strength of the exposure-confounder relationship. In the majority of situations examined, however, dependent samples were found to be somewhat more efficient than independent samples when confounding was present, while the reverse was true when confounding was absent. Results of small sample simulations do not differ importantly from the asymptotic results, except for pair-matching on a non-confounder, where the inefficiency of matching is greater in small samples.

Sample size determination in case-control studies: The influence of the distribution of exposure

In the design of case-control studies, the determination of the required number of cases and controls is based on consideration of the strength of the relationship between the disease and exposure to the putative cause, the variability in exposure within the population under study, and the desired size and power of the statistical test. This paper derives sample size equations for studies with a continuous exposure which allow the investigator to specify the strength of the relationship between disease and exposure in terms of relative risk, etiologic fraction or the slope of an exposure response relationship. With these formulations it becomes apparent that the size of the sample increases with decreasing variability in exposure in the population under study, a finding not apparent in the use of conventional methods of sample size determination for continuous data. The ability of a case-control study to detect a significant association between disease and exposure therefore depends critically on the distribution of exposure which exists in the community to be studied. Implications of these findings for studies of diet and cancer are discussed.

Physician Sensitivity to the Price of Diagnostic Tests: A U.s.—canadian Analysis

A questionnaire containing 11 patient management problems was completed by 495 physicians and medical students at an American and a Canadian medical school. Respondents indicated whether they would order a particular diagnostic test in each case, given five different prices for the test. Approximately 25 per cent of attending staff and a higher proportion of residents, interns and clerks responded that they would order the test depending on its price. Approximately 50 per cent of attending staff and smaller proportions of residents, interns and clerks indicated that they would not order the test even if there were no price. Respondents in Montreal were more likely than those in Philadelphia to select a price-sensitive response, the reverse of the expected tendency. Since some tests may be ordered on the basis of price, education of physicians regarding the price of diagnostic tests may alter their use of these services, but a large proportion of tests are ordered because of clinically absolute reasons, which may be insensitive to price.

Inter-observer error among surgeons and nurses in presymptomatic detection of breast disease.

Abstract One hundred and three women were examined independently for presymptomatic breast disease by two nurses and two surgeons who recorded physical findings and their recommendations for further clinical workup. Agreement between the observers beyond what would have been expected by chance was assessed by a new extension of the statistic κ which allows multi-level scales of measurement, more than two observers (not necessarily the same for each subject), and comparisons between and within subsets of observers. Agreement between nurses and between the nurse-surgeon pairs was not significantly better than would have been expected by chance. Agreement within surgeon pairs was only slightly better than chance (overall κ for physical findings and recommendations being 0.42 and 0.32 respectively). Agreement between surgeons was generally better for physical findings than for recommendations and was best for the finding of fibrocystic disease. Future studies to compare the performance of nurses or other allied health professionals with surgeons should, therefore, be designed to allow assessement of the reliability of the standard group.

Measuring the impact of an intervention on equipment lives

A 1967 strike at a Quebec aluminum smelter resulted in the uncontrolled shutdown of aluminum-reduction cells in the smelter’s potrooms. In a subsequent legal action against the union which was before the courts for more than a decade, the company claimed that the shutdown had reduced the operating lives of the hundreds of cells in service at the time. This study describes the background and outcome of the court case and presents the data used by expert witnesses to argue for and against the company’s claim. Our analysts independently examine the data and arrive at their own conclusions.