| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
EDITORIAL COMMENT |
Stanley Cobb Professor of Psychiatry, Harvard Medical School, Director, Harvard Institute of Psychiatric Epidemiology and Genetics, Superintendent and Head, Harvard Department of Psychiatry, Massachusetts Mental Health Center, 74 Fenwood Rd., Boston, MA
In this issue of Psychosomatic Medicine, Christensen et al. (1) presents the results of their study that examined the role of social support as a predictor of long-term survival among patients with schizophrenia. Their analyses indicated that a higher quantity of social support available to the schizophrenic patient at or before the time of psychiatric admission had a positive impact on subsequent survival time.
By using a retrospective cohort design, they examined archival patient records for a group of schizophrenic patients admitted to a psychiatric hospital between 1934 and 1944. All of the patients were deceased at the time of an average follow-up period of 58 years. Each patient’s social history, as recorded at the time of admission, was abstracted from the medical record and evaluated/scored by two independent raters, for quantity and quality of social resources available to the patient. With the use of this information, the authors tested the hypothesis that either the quality or quantity, or both, of social support would have a positive impact on longevity.
This type of design (retrospective cohort) is known as a nonexperimental (observational) study, where the investigators have no control over the group designation of each study subject (2). The researchers generally select subjects for the different exposure conditions (ie, social support) from previously existing groups (ie, hospital-admitted schizophrenics), and then observe the resulting health outcomes (ie, survival times). Even though this is a cost-effective design, there are a number of limitations associated with this type of study. One limitation involves the quality of data regarding exposure. Because the researcher has no control over data collection he or she relies on existing record information. This is the case in the article by Christensen et al. (1) where the determination and measurement of the exposure variable (social support) relies totally on medical record information. Other problems concerning follow-up in retrospective cohort studies do not apply to this particular study, because all of the cases were deceased at follow-up.
Our long-term follow-up and family study of schizophrenia, mania, and depression (3) likewise used a retrospective cohort design. Even though our study also used medical records for patients admitted to the Iowa Psychopathic Hospital between 1934 and 1944 to define our study groups of schizophrenia and affective disorders, we performed a 30- to 40-year follow-up after discharge, by training individuals to personally interview patients and their family members by using a structured diagnostic instrument. The process of examining outcome, mortality, and family data for the various diagnostic groups, as defined by medical record information, provided a mechanism to evaluate the accuracy of the original record data.
Under an ideal study design, would Christensen et al. (1) obtain the same results and conclusions that they have presented here? In a perfect world, the investigators would have a choice of study designs. In addition to the observational design used, they also could have chosen an experimental model. In an experimental study, the investigator controls the allocation of the subjects to different comparison groups and also regulates the experimental conditions of each group. Study subjects are randomly assigned to comparison groups and followed up over time to record the outcome event of interest. Clearly, if the researcher is able to conduct either an experimental or an observational study, he or she would prefer experimental studies, inasmuch as one could then control the conditions under which the study is conducted. In the above-mentioned study, in which the authors are testing the hypothesis that social support in the early stages of schizophrenia will improve the length of survival, they could have theoretically used an experimental design. In this design schizophrenics would randomly be assigned to two or more well-defined social support groups and then followed up over time to examine the impact on survival.
I believe there are a number of points that many researchers and readers of journal articles have lost sight of when they examine the specific design of a particular study. Epidemiological research is based on two levels of reasoning: a) the determination of a statistical association between a characteristic and a disease; and b) the derivation of biological inferences from such a pattern of statistical associations (4). Consideration of these factors helps us answer the questions: Are the results of the study biologically significant and is the research design adequate to test the statistical associations? The demonstration of a statistical relationship between a disease and biological or social characteristics is only the first step in epidemiological analyses. The second step is to ascertain the meaning of the relationship. Another goal of the investigator is to determine whether an etiological hypothesis developed clinically, experimentally, or from other epidemiological studies is consistent with the epidemiological characteristics of the disease in the human population group.
Christensen et al. (1) do a good job in addressing some of these issues in their Discussion section. For instance, they describe a possible model to explain an underlying biological mechanism. They state "there is considerable evidence that social support influences cardiovascular, immunological and neuroendocrine responses." Secondly, they address the sampling problems by adding that "causal interpretations of the present data should be made cautiously, due to the correlational and archival nature of the study methodology." Finally, the authors allude to the possibility of using an experimental design by stating that "more compelling causal evidence may require research examining the effects of manipulated social support on patient outcomes." By summarizing the above in the Discussion section, they have thoughtfully considered many of the methodological issues I have raised in this editorial.
Readers should consider two very important points when they read research articles: a) Is the study design appropriate for the hypotheses being tested and what are the limitations of that particular design? and b) Does the sample selected for the study represent the corresponding group in the general population? These are fundamental to performing good research, and should never be taken for granted.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
