Analysis of Clinical Follow-up Databases: Risk Stratification Studies and Prospective Trial Design MAREK MALIK From the Department of Cardiological Sciences, St. George's Hospital Medical School, London Malik, M.: Analysis of Clinical Follow-Up Databases, Risk Stratification Studies, and Prospective Trial Design. Design of new prospective studies should utilize detailed retrospective evoluotions ofcHnicaldata. For this purpose, clinical data ore needed containing both prospectively recorded values of risk factors and follow-up events. A concept of a new trial can then be modeled within the existing data-set. The de- velopment of such a model consists of the following steps: (a) the distribution of values of risk factors has to be investigated in the whole recorded population and the statistical association of the risk factors with follow-up events has to be established; (b) the stratification characteristics (sensitivity, specificity, and predictive accuracies) have to be evaluated for individual risk factors and for their multivariate combina- tions; (c) the stratification characteristics have to be converted into estimates of mortality reduction ex- pected within the high risk group and used for the optimum trial design in terms of screened and ran- domized patient numbers. In this review, the strategy of designing a new trial is demonstrated using data of 644 survivors of acute myocardiai infarction with available 3-year follow-up during which 74 patients died, A mode! of a new trial is considered involving reduced left ventricular ejection fraction, increased 24-hour mean heart rate, and depressed 24-hour heart rate variability as risk stratifiers. (PACE 1997; 20(Pt. 111:2533-2544) Introduction For rather obvious reasons, the appropriate- ness and efficacy of every novel treatment and, in particular, every novel prophylactic option should be validated and tested in a prospective study. However, reasonable design of new prospec- tive studies should utilize detailed retrospective evaluations of clinical data. Experi- ence shows that the goals of prospective trials based on traditional beliefs and consensus assump- tions are far less frequently fulfilled than the goals of prospective studies the design of which was ret- rospectively modeled within existing data sets. In- deed, the hypothetical concepts derived only from speculations about the contemporary state-of-the- art knowledge might be easily misleading and in the past, many such concepts were used in propos- Address for reprints; Marek Malik, PhD, MD, Department of Card io logical Sciences, St. George's Hospital Medical School, London SW17 ORE, United Kingdom. Fax: +44-181-767-7141; E-mail: m.malik^Jsghms.ac.uk Received February 10, 1997: revised April 30. 1997; accepted April 30, 1997. als of new prospective studies although data were available which, if properly analyzed, would clearly show that the new proposal is unrealistic. Modeling a proposal of a new prospective study in an existing data set is, of course, based on the assumption that the nature and character of data accumulated in the past will be repeated in the future. While this is rarely the case, a retrospective modeling test is certainly much more accurate than a hypothetical consideration. As the methodology of performing a retrospective test within an existing clinical follow-np database is, in some aspects, rather novel, this article is aimed at summarizing the individual steps which should be considered when organising such a test. Example Data In order to demonstrate the individual con- cepts and data analytical approaches, follow-up data accumulated in the database of the Post Infarc- tion Research Survey of St. George's Hospital will be used. In particular, the examples in the text will use a population of 644 patients (mean age 56.5 ± 8.9 years, 131 women) who suffered from acute my- PACE, Vol. 20 Octoher 1997, Part II 2533