Evaluating the reliability of causes of death in published clinical researchBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7076.271 (Published 25 January 1997) Cite this as: BMJ 1997;314:271
- R D Start, lecturer in pathologya,
- J P Bury, medical studenta,
- J P Strachan, medical studenta,
- S S Cross, senior lecturer in pathologya,
- J C E Underwood, professor of pathologya
- Correspondence to: Professor Underwood
- Accepted 24 September 1996
Deaths during clinical trials or outcome review studies could be due to the disease being studied; adverse complications of interventions in the study; intercurrent disease unrelated to the initial condition or interventions; or unnatural events. Accurate ascertainment of the causes of deaths is therefore required to establish whether treatment has prevented deaths from the investigated disease without concomitant increases in death or morbidity from other causes. We assessed the methods used to evaluate causes of death in published clinical research.
Methods and results
All 879 papers published in 1994 in the Lancet, BMJ, Annals of Internal Medicine, and New England Journal of Medicine were reviewed; the methods used to evaluate the causes of death in each paper were recorded using 11 categories (table 1). Death was an outcome measure in 223 papers (25%). The two largest categories of papers were those in which causes of death were either unstated or stated without explanation of the method of ascertainment. The most common stated methods were clinical review committees and registries or databases. Forty two papers (19%) used one or more methods. A necropsy was used in only 29 papers (13%), and the proportion of deaths examined by necropsy varied widely (mean 60%; range 25-100); 15 of these papers failed to include the proportion that underwent necropsy. Four papers were based entirely on necropsy data.
This survey revealed considerable variation in the methods used to evaluate the causes of death in clinical research. Many publications contained no information about the causes of deaths. Some data may have been omitted through space constraints and editorial processing, but the methods used to ascertain causes of deaths should be stated in all clinical research publications, particularly when so few studies verify more precisely by necropsy the role of interventions in preventing and causing deaths.
Randomisation is insufficient to permit the assumption that differences between intervention groups in causes of death must be due to the intervention. Although randomisation should ensure that groups are similar, accurate ascertainment of the causes of deaths is still required to establish whether interventions have prevented deaths from the investigated disease without concomitant increases from other causes, including serious adverse events. Apparent disease associated mortality may actually be due to serious complications of treatment; thus a major contribution of a necropsy is the identification of both non-lethal and lethal side effects of medical interventions.1
The reasons for low necropsy rates in clinical research are complex and include inability to obtain relatives' consent and difficulties in identifying subjects in clinical trials at the time of death, particularly when it occurs outside hospital. Large multicentre studies may encounter additional problems through local variations in medicolegal practices, necropsy request procedures, and cultural and religious attitudes towards death and necropsy. Although a detailed postmortem examination may not always identify a definitive cause of death, any death which is believed to be related to medical treatment, experimental or otherwise, should be reported to the appropriate medicolegal authority for further investigation. However, medicolegal necropsy data are often unavailable to investigators.
Clinical review committees are commonly used to evaluate causes of death and, when necropsy data are also available, are the most reliable method.2 Nevertheless, there remains considerable reliance on data from death certificates, which are often inaccurate.3 Those concerned in performing clinical research should be aware of the limitations of methods used for determining the cause of death. Methods used in individual studies should be clearly stated, and strenuous efforts should be made to include necropsy data. Recent initiatives to improve the reporting of clinical trials are timely,4 5 and all clinical research in which deaths are expected or represent outcome events should include protocols for evaluating deaths. Ideally, this should be by necropsy in conjunction with clinical information, but we acknowledge that many countries have low necropsy rates and it would not be feasible to make necropsies mandatory.
Conflict of interest: None.