Introduction in order to study the association between disease status (eg, nosocomial infection) and some exposure variable (eg, number of days on urinary catheter), it is often necessary to take into account other variables that may influence either the disease status or the exposure variable. For example, Freeman et al described a retrospective (in their terminology “case-referent”) study of a neonatal population. In this study, cases of nosocomial infection are selected in addition to corresponding control individuals who did not experience a nosocomial infection. All patients were selected from a neonatal intensive care unit, and the goal was to study the role of umbilical artery catheterization and its association with nosocomial infection. As noted by Freeman et al, this is a complex question because the risk of nosocomial infection might reasonably depend not only on the duration of catheterization, but also on the birth weight of the infant.

Previous papers in this series have discussed the design of epidemiologic studies and analysis for data that can be presented in a 2 × 2 table. The purpose of this paper is to explain how sample sizes are determined for unmatched prospective and retrospective studies. One of the most common questions asked in planning a project is “What sample size do I need for my study?” Determining the correct sample size is an important consideration that is best resolved prior to data collection. In this paper we assume the goal of the study is to examine the association between a dichotomous risk factor and the presence or absence of disease; therefore, the analysis will be that of a 2 × 2 table. We also assume throughout that sample sizes are to be equal for the two comparison groups (ie, equal numbers of exposed and unexposed for a prospective study, and equal numbers of cases and controls for a retrospective study). Sample size curves used to determine sample sizes will be presented. The results here are based on a paper by Schlesselman that contains sample size formulae for these study designs, and to which the reader may refer for an excellent technical discussion of their derivation.

Clinical epidemiologic studies often focus on the identification of risk factors associated with hospital-acquired infections and the subsequent effects of preventive measures. Two study designs commonly used are the prospective design and the retrospective design. Both of these designs and the associated data collection issues were defined and discussed in the first article of this series. This article reviews the statistical analysis for a dichotomous risk factor and the presence or absence of a given disease for unmatched data. We focus our attention primarily on the unmatched prospective study, although we present formulae for the retrospective study as well.

Much of clinical and hospital epidemiology involves the identification and enumeration of cases or the comparison of case frequencies between two or more groups of interest. Because both of these activities involve the use of statistics, it is important to pay careful attention to biostatistical issues involved in the collection and the analysis of such data.

In this article, the first in a series of biostatistical papers, we discuss some general issues that are important in the design, analysis, and interpretation of clinical epidemiologic data. Subsequent papers in this series will deal with specific methods of analysis, examples of these methods of analysis, and limitations and interpretations of the methods.

Schizophrenia and affective disorders selected according to the Feighner criteria can be differentiated on the basis of 40-year outcome. Within schizophrenia the presence of disorganized thoughts at the index admission was associated with poor outcome, whereas better outcome was associated with the presence of delusions or hallucinations. Within the affective disorders, bipolar patients with grandiose delusions or ideas showed a poor outcome; a better outcome was found in unipolar patients with complaints of fatiguability or tiredness at the time of the index admission.

Causes of death were studied in a cohort of 200 schizophrenic, 100 manic, and 225 depressive patients who were followed in a historical prospective study. These patients were admitted between 1934 and 1944 and were studied 30 to 40 years later. Five cause of death categories were considered in this analysis: (1) unnatural deaths, (2) neoplasms, (3) diseases of the circulatory system, (4) infective and parasitic diseases, and (5) other causes. For each cause of death, the expected number of deaths was calculated from vital statistics for the State of Iowa for the time period of follow-up. Observed numbers of deaths were contrasted with expected numbers of deaths to assess statistical significance for each diagnostic group. There was a significant excess of unnatural deaths in all diagnostic groups in both sexes, with the exception of female manics. This group, however, did show a significant excess of circulatory system deaths. Both male and female schizophrenics showed a substantial excess of infective disorder deaths.

Mortality data are presented from a four-decade follow-up study of 200 schizophrenic, 100 manic, 225 depressive patients, and 160 surgical controls (80 appendicectomy; 80 herniorrhaphy). Data for this analysis were available on 648 (95 per cent) members of the study population. Using sex-age standardized mortality ratios (SMR), the mortality experience of the study population was compared with that of the state of Iowa, the geographical area served by the admitting medical facility for the study group. Results are presented for a four-decade period beginning 1935–44, and ending 1965–74. All three psychiatric groups had a significant increase in mortality risk. This was most pronounced in the first decade following admission, although schizophrenic patients, especially females, continued to show a significant excess of deaths throughout the entire four decades of the follow-up period. During no decade of the follow-up period did the mortality of the surgical controls differ significantly from that of the Iowa population.