The quality of medical death certification of cause of death in hospitals in rural Bangladesh: impact of introducing the International Form of Medical Certificate of Cause of Death

This study was conducted as part of a National Health and Medical Research Council (NHMRC) multi-country research project. Initially, the study investigator discussed project objectives with senior hospital staff members. The death certificate used by public hospitals was not structured like the international death certificate, so the core components/parts of the international MCCD were incorporated, without affecting any information available on the present death certificate. Accordingly, a death certificate was prepared and supplied in every unit of the hospitals where death certificates were issued. Several workshops were organized to train 250 doctors how to complete the death certificates before starting the project activities. Through didactic and interactive lecture, the workshops oriented doctors and nurses about the principles of death certification and the requirements of good clinical record keeping for quality data. Trainees were provided with practice clinical cases to assign a COD. Periodic refresher courses with doctors and interns were also conducted by study physicians to improve certification practices. A booklet on cause of death certification was distributed among all doctors, and a laminated guideline was displayed in each doctor’s room [13]. The project staff compiled completed death certificates and photocopied the clinical records of the deceased including available laboratory investigations for subsequent analysis.

Two study physicians were briefed and trained about the different sections of the medical data extraction forms and gold standard criteria for diagnoses of main causes of death among neonates, children, and adults. They were trained how to review the MCCD and clinical records to complete the Medical Data and Audit Form (MDAF). They were also trained on the ICD-10 manual.

The study took place between January 2011 and April 2014 in the Chandpur and Comilla districts of south-eastern Bangladesh. The study initially included all 8 public hospitals (one secondary level Chandpur District hospital and 7 primary level Upazila Health Complexes) and 6 private hospitals (including the International Centre for Diarrhoeal Disease Research (icddr,b) hospital) operating in the district headquarters of Chandpur district. In 2011, only 404 deaths were identified in all of these hospitals which provided about 500 beds for a population of approximately 2.3 million. The study was thus expanded to the Comilla Medical College Hospital (tertiary level) which has 500 beds, a daily overnight hospital stay of 750, and helps serve a District population of 5.3 million.

All deaths in the study period that occurred in the selected hospitals and for which death certificates and medical records were available were included. Medical records and death certificates were photocopied for study purposes. Deaths were classified as neonates (first 27 days of life), children (28 days – 12 years) and adults (12 years and older).

Bangladesh routinely reports hospital deaths from a single line death certificate and uses these reports to describe hospital mortality nationally. We introduced the international death certificate, which has three sections:

The doctors were instructed to assign their diagnosis according to international rules which included 1) when there was only one cause identified, they would put them in the first line of part 1 of the death certificate 2) when two or more causes were identified, they would assess the causal relationship between them, and they would write the causally related causes in part 1 with the underlying cause on the lowest line. They would write the cause which contributed but was not causally related to main causes into part II of the death certificate as the contributory cause.

The study physician reviewed the hospital death certificates and clinical records to rewrite a new death certificate and allocate an underlying cause of death to the case for the purposes of this study. While doing this, the study physician was instructed to note whether they assigned a new underlying cause or just made a change in the order of diagnoses on the hospital death certificate.

The study developed a Medical Data and Audit Form (MDAF) that included 1) basic information of the hospital and the deceased 2) the same cause of death information from death certificate (verbatim) 3) cause of death from medical record review by the study physician 4) quality of clinical records – do the clinical data justify the clinical diagnosis; was it necessary to change the clinical diagnosis or sequences in the diagnosis to allocate the UCOD? The quality of medical record review-based diagnosis was assessed according to gold standard diagnosis criteria used by the Gates 13 Grand Challenges in Health (GC13) study. This study used the gold standard criteria developed by the Population Health Metrics Research Consortium (PHMRC) [14]. These gold standard criteria were developed by a committee of physicians and underwent multiple cycles of group review. The gold standard criteria classified the deaths into six levels based on the degree to which the information from medical records provided certainty to classify the death for a given cause: level 1) Gold standard 1 (GS1): highest level of certainty- diagnosis of a particular condition with the highest level of certainty possible for that condition, consisting of either an appropriate laboratory test or x-ray/imaging with positive findings and/or medically observed and documented appropriate illness sign(s) to a pre-determined standard. 2) GS2A: high level of certainty – diagnosis of a particular condition with a high level of certainty, consisting of appropriate laboratory/x-ray with positive findings and /or medically observed and documented appropriate illness or signs to a pre-determined standard. 3) GS2B: high level of certainty – presumed initial diagnosis of a particular condition with high certainty; this category was kept reserved for cancer and HIV patients on long-term treatment where initial data had been lost. 4) GS3: Reasonable level of certainty – medical diagnoses not supported by appropriate level of lab investigations but which meet established clinical criteria. 5) GS4: Unsupported – medical diagnoses not supported by the adequate observed and documented clinical evidence/criteria. 6) Other – ranking does not have an appropriate GS category.

To identify the likely leading causes of death in Bangladeshi hospitals we categorized causes of death from study death certificates. Included death certificates had a valid age, sex, rank of quality, and study ICD-10 code. Age was categorised as neonate, child, and adult categories. The number of times the study physicians had to change the underlying cause of death or the order of diagnoses on the hospital death certificate was tabulated and stratified by age, hospital, and for the top 5 most frequent gold standard causes of death for adults, children, and neonates. The gold standard underlying COD was given as an ICD-10 code by the study physicians. The top 15, 10, and 3 most frequent ICD-10 codes for adult, children, and neonates, respectively, were given a text COD corresponding with the ICD code. All other ICD codes were given a text COD corresponding to the ICD chapter name (see Additional file 1). The frequency of GS levels was stratified by the same categories.

The frequency of the top 20 causes of death for adults, top 10 causes for children, and top 5 causes for neonates were tabulated. Gold standard COD given as an ICD-10 code was translated to text using the same mapping methodology, but with causes of death outside the top 20, 10, and 5 most frequent causes of death for adults, children, and neonates collapsed into the “other” category.

Common hospital death certificate errors were tabulated for each death certificate: multiple causes per line, change in diagnosis necessary, change in sequence necessary, interval between onset and death not shown, abbreviations used in certifying death, leaving blanks between lines in part 1, ill-defined conditions entered as underlying cause of death, and ambiguous poisoning deaths. Since hospital physicians tended to use “with”,“&”, “and”, “as a result of”, and “due to” delimiters when listing multiple causes of death on the same line, these delimiters were used to identify death certificates that included more than one cause of death on the same line in part 1. Death certificates were determined to have an interval between onset and death if every contributing cause of death had a corresponding interval. The presence of abbreviations on each line of each death certificate were tabulated by creating a subset that contained two or three letter words or known four letter abbreviations (“COPD”, “VLBW”, etc.). Common English words (“to”, “and”, etc.) were then removed. Death certificates that had a blank between lines in part 1 were tabulated. The first line of death certificates that contained the ill-defined keywords “failure,” “septicaemia,” “shock,” “cardio,” “brain,” and “old” were manually examined. Lines that did not include any other specific information were deemed ill-defined UCOD. Poisonings that did not indicate the type of poisoning or whether the death was accidental or suicidal were tabulated. The frequency of death certificates with any of the above errors or without an immediate cause of death (entry in the first line) was calculated. The frequency of any error except those that did not include a time interval was also calculated, as was the frequency of common abbreviations by age group.

Only 6.7% of neonate medical records were of high quality, compared to 38.5% of adults and 21.6% of children (Table 2). Quality of medical records was highest for adult stroke (57.9%) and child gastroenteritis and colitis (48.6%) (Table 3). Quality of neonate and child medical records by cause of death was lower compared to that of adults.

It was necessary to change UCOD for 45.5% of death certificates, with little difference by age group, sex, hospital, and cause of death (Table 4, Additional file 2). Child deaths due to gastroenteritis and colitis required more changes in the UCOD (60.0%) than all other causes for adults, children, and neonates. For most death certificates, changes in the UCOD were between 2 and 4 times more likely to have occurred due to a change in sequence than a change in diagnosis.

For the types of death certificate errors, nearly all physicians (95.6%) did not indicate the time interval between onset and death, 31.6% required a change in sequence, 13.9% required a change in diagnosis (Table 5). 99.1% of death certificates had any error, but 87.8% had any error other than the interval between onset and death not shown. The most frequent abbreviation for adults was “MI” (myocardial infarction) (9.8%), for children was “BR.” (breathing) (13.7%), and for neonates was “PNA” (perinatal asphyxia) (39.4%). No death certificates indicated neoplasm as a COD, and injuries were omitted from the error analysis because nearly all injuries did not indicate an intention.

For adults, the top three causes of death were stroke (15.8%), acute myocardial infarction (14.4%), and other cardiovascular diseases (5.9%) (Table 6). Adult deaths also had a relatively large proportion of deaths that were “impossible to specify” (3.1%). For children the top three causes of death were pneumonia (31.7%), sepsis (10.4%), and gastroenteritis and colitis (7.6%) (Table 7). For neonates, the top three causes of death were birth asphyxia (52.8%), sepsis (20.9%), and low birth weight (15.2%) (Table 8).