Methods of multicenter trials in psychiatry part I: Review

Abstract

General

Fifty out of 100 publications on multicenter trials and 21 on methods are listed and partly discussed. Our discussions concentrate only on double blind, therapeutic trials. Multicenter trials are the only way, to keep sources of heterogeneity or error under control. Making high demands, the advantages of well-done multicenter trials surpass by far the disadvantages.

• 1.

  1. How to conduct a multicenter trial.

  • 1.1.

    1.1 Introduction: In a trial the variabilities of interest are the efficacy and tolerance of drugs. Therefore all efforts have to be made to minimize all other non-treatment variances or to keep them constant. A multicenter trial has to be a logical, feasable, carefully planned and standardized sequence of events, carried out exactly as planned.

  • 1.2.

    1.2 Sample consideration: Heterogeneity of patients' diagnoses is an additional variance. Every variability beside the drug-effect will lower the success of the trial, the reliability and validity of the results. To prove or reject a hypothesis, the only target patients for an efficient proof are endogenous depressions (antidepressant), exacerbated paranoid schizophrenia (neuroleptic) and chronic anxiety states (minor tranquilizer) of medium to severe degree. Despite difficulties diagnostic labeling of patients (WHO-ICD 9) is an absolute necessity: symptoms are only meaningful in the context of the diagnosis concerned. Also the course of illness, past and present, is of importance (spontaneous remission during study). Statistical reasons require much higher sample sizes than are usually considered: with an $\text{α}$-risk of 5 %, a $\text{β}$-risk of 10 %, a 50 % expected improvement of a standard and a 60% improvement of a new drug, two times 422 patients are necessary to demonstrate a difference.

  • 1.3.

    1.3 Settings and investigators: Cross-study variabilities of numerous sources should be minimized as much as possible, concentrating on efficacy and tolerance of drugs: “prevention is better than cure”. The higher the qualification, experience and capability of investigators and nurses, the better will be the results and reliability.

  • 1.4.

    1.4. Experimental design: Clinical multicenter trials should be organized and conducted in a way which resembles ordinary clinical practice and being in the best interest of the (future) patients (Declaration of Helsinki/Tokyo). The design, the protocol, patient forms, execution and evaluation should be a master-piece of clarity. There is only one way to minimize undesired variability and deviations: standardization of every detail from the beginning to the end, considering logical thinking, reality, feasibility and practicability. One kind of standardization should never be attempted, a fixed dose of drugs. Treatment duration should not be less than 6 weeks (phase I and II).

    Up to now nearly all rating scales frequently used have been based on psychological thinking. We need scales built on clinical, psychiatric and statistical experience, with high practicability and a broad spectrum of the most important, unbound symptoms. Interrater-reliability test data of scales are mostly related to highly trained raters under maximal conditions but not — which is necessary — to normal practical situations: each rater population has its own reliability. Of much more importance is the intrarater-reliability (consistency, stability) e.g. in untrained raters. Content-validity is dependent on the clinical/psychiatric input. Construct-validity can be demonstrated by factor- and cluster-analyses and can be proved by the clinical relevance of their results. But concurrent validity, opposing a new to a known rating scale is like comparing the blind and the deaf. A scale has to be compared to evident clinical parameters like global therapeutic efficacy.

  • 1.5.

    1.5. Execution and documentation: In this phase quality assurance maintains or enhances the reproducibility and validity of study findings. A good quality-consciousness and a quality assurance system should cover all aspects of the data generation and analysis process since missing data, sloppy data processing and analysis procedures can be at least as deleterious to end results as carelessness in the generation of data. This includes not only (basic data) dosage and efficacy data but also untoward effects, vital signs and laboratory testing.

  • 1.6.

    1.6. Analysis, synthesis and interpretation: The systematic involvement of a biostatistician from the very beginning (planning) to the end (statistical analysis) is a “sine qua non” for any clinical study, especially for a multicenter trial. The statistical advice helps to keep down all undesired variables and to obtain data which can, without loss, undergo statistical analysis.

    “Obviously it is important to protect the investigation per se from the major hazard of inappropriate criteria, inappropriate samples, and the numerous sources of error and confounding in the treatment of the patient and in the collection of the data”. A multicenter trial should be stratified a priori (separate randomization of single studies) and be analyzed separately. Differences among hospitals and investigators should be identified (including interaction tests) before pooling of all data can be arranged. If a quantitative combination (pooling) of single study results is not possible, there are several methods for combining data qualitatively (For statistical methods see sourcebooks).

    The interpretation (final report) has to present the single studies and the pooling separately. The main principle for this (between drugs, populations, studies and for the pooling) is the description of all details concerned: a) the initial (predrug) comparability; and b) the comparability in course of time (treatment period); as well as c) the final comparability.

    The final report should reflect the standard procedure of the studies and multicenter trial in structure and content so as to facilitate the understanding and the comparability to practice. It should be possible in a well documented and well presented study to trace an individual patient's raw data through to it's contribution in arriving at a probability statement (computer outprints, tables and listings). The aim is a final monograph in which all the data of the studies and the multicenter trial are combined. Such monographs have a long standing value.

• 2.

  2. Conclusions: Open and double blind multicenter trials are the only way in psychopharmacological research to achieve the large sample sizes which are, for statistical reasons, needed. Prerequisite for meaningful statistical and clinical results is to minimize non-treatment variables. There is one way to succeed: uniform, standardized, strict and well controlled conditions and procedures, cooperation with a biostatistician, slavish attention to a myriad of details, quality consciousness and control from beginning to end. But we never should forget the primary goal of all our science and research: the benefit of the patient.