Studies of prognostic factors in end-stage renal disease: An epidemiological and statistical critique

We reviewed prognostic studies for patients treated with renal replacement therapy by using an electronic database and bibliographic review for 1990 to 1998. Using the inclusion criteria of English language, adult patients, primary article, minimum 50 patients, primary focus on prognostic factors, and mortality outcome, 104 articles were identified. The 104 articles were reviewed for eight epidemiological and seven statistical criteria that addressed the scientific validity and interpretability of results. The following percentages of the 104 articles satisfied each of the eight epidemiological criteria: (1) a priori hypothesis, 6%; (2) zero time specified, 49%; (3) prognostic factors collected before zero time, 69%; (4) inception cohort, 59%; (5) control for treatment, 74%; (6) operational criteria, 82%; (7) missing variables reported, 12%; and (8) loss to follow-up reported, 42%. Summary analysis showed that 76% of studies satisfied four or fewer of the eight identified criteria. In the 77 articles (74%) that used the Cox proportional hazards model, the following percentages of articles met each of the seven statistical criteria: (1) proportional hazards verified, 26%; (2) censoring explained, 57%; (3) multivariate analysis performed, 91%; (4) significance levels given, 99%; (5) age adjusted, 95%; (6) diabetes adjusted, 66%; and (7) cardiac adjusted, 44%. Summary analysis found that 47% of the 77 studies satisfied four or fewer of the seven identified criteria. Superficially, results appear to show that when the Cox proportional hazards model was used, statistical analysis was better than the epidemiological design. However, studies we examined had serious defects in both epidemiological design and statistical analysis. The consequent validity of results for the quantification of prognostic factors is questionable.