Bulletin of the World Health Organization

Impact of misclassification on measures of cardiovascular disease mortality in the Islamic Republic of Iran: a cross-sectional study

Ardeshir Khosravi a, Chalapati Rao b, Mohsen Naghavi a, Richard Taylor b, Nahid Jafari a, Alan D Lopez b

Introduction

Measures of mortality, such as age-specific death rates, life expectancy, cause-specific death rates and years of life lost, are commonly used to measure the health status of a population and are essential for epidemiological research and priority setting for health development.14

In general, countries can be classified into two broad groups on the basis of availability of data on causes of death.3 One group comprises countries that typically have complete vital registration with medical certification of the cause of death assigned by attending physicians. The other group includes countries that have death registration systems ranging from incomplete to virtually non-existent, where causes of deaths are often recorded inaccurately at registration, resulting in large proportions of deaths assigned to ill-defined causes.1,3 In some countries a history obtained from relatives or associates, known as “verbal autopsy” or “lay death recording”,5 is available for some segments of the population. In China and India, for example, sample vital registration areas that collect information via the use of verbal autopsy provide useful and representative information on causes of death.6

Despite international efforts to facilitate and standardize processes for the collection and coding of data on causes of death, the quality of data from many countries remains poor.13 Several factors influence cause of death ascertainment, such as the nature of the disease or circumstances of death, the qualifications and skills of the certifier, and the availability of diagnostic aids and medical evidence. Variations in these factors probably contribute to significant misclassification of cause of death in many countries.7,8

Studies to assess the accuracy of data on causes of death from routine death registration systems have been implemented in several countries.4,7 In general, these studies compare causes of death reported on death certificates with diagnoses from clinical records or autopsies. These studies can only be conducted in populations for which detailed clinical records (or autopsies) and data from death registration systems are available and useable, as in the Islamic Republic of Iran.

Death registration in the Islamic Republic of Iran was initiated by the National Organization for Civil Registration in 1918, and has evolved over the past few decades into a new comprehensive death registration system operated by the Ministry of Health and Medical Education (MOH&ME).914 Starting in one province (Bushehr) as a pilot study in 1997, the new system was progressively implemented to cover 29 out of 30 Iranian provinces by 2006.10,1517 In urban areas, attending physicians complete a medical certificate of cause of death in accordance with the principles of the International Classification of Diseases (ICD). In rural areas, causes of death are determined by physicians from the local rural health centre, based on “verbal autopsy” interviews conducted by health-centre staff.15 Each month, urban and rural health facilities submit summary information on age, sex and up to three causes for each death to the district health centre, where the data are matched against other sources (e.g. cemetery, hospital or medico-legal records) to improve the level and quality of death registration, and to remove duplication. Next, trained personnel select and code the underlying cause for each death according to ICD rules. These data are then computerized and submitted to the provincial health department, where data are cross-checked with information from the National Organization for Civil Registration to further reconcile missed deaths or duplications. A final dataset is submitted to the MOH&ME where deaths are tabulated according to an abbreviated list of 321 causes based on ICD-10, but adapted to represent the epidemiological profile of the Islamic Republic of Iran.15

A perusal of leading causes of death from registration data (Table 1) raises considerable concerns about their utility, with five being nonspecific or vague cause categories such as “senility without mention of psychosis”, “unknown”, “other cardiac diseases”, “other unspecified disorders of the circulatory system” and “other respiratory diseases”. Further, although “heart failure” and “hypertensive diseases” are recognized causes of death, the number of deaths classified to these two categories appears disproportionately large in comparison with other countries with good-quality data.18 The validity of reported data for these two specified causes should therefore be assessed to ascertain whether these proportions are true, or are an artefact of death-certification practices. Similarly, an assessment of the “true” causes of those deaths classified to the five nonspecific or vague categories could help identify the principal patterns of misclassification to these categories. This study addresses these two specific issues by comparing registration diagnoses with reference diagnoses derived from medical records for a sample of deaths that occurred in health facilities in the Islamic Republic of Iran during 2005.

Methods

Study design

For the purpose of measuring validity of registered causes of death, a reference diagnosis for each death in the study sample is required for comparison. In view of the limited availability of pathological autopsies as “gold standard” reference diagnoses in the Islamic Republic of Iran, medical records might serve as a suitable alternative.8,19 Analysis of death registration data from 2003 and 2004 indicated that about 45% of deaths in the Islamic Republic of Iran occurred in hospitals, suggesting that hospital medical records might constitute the best source for reference causes of death for this study.15,20 To assess the feasibility of using this information, a pilot study was conducted on a random sample of 100 deaths from one hospital in each of two provinces (Kermanshah and Yazd) in 2005. The aims of the pilot study were to review the availability of medical records in each hospital, assess their suitability to establish reference diagnoses, and estimate the resources required to do so. Based on the pilot study, it was apparent that the quality of medical records might well vary substantially between provinces. Furthermore, the quality of medical records for some deceased persons who died at an advanced age, or who had a short length of stay in hospital, was generally very poor.

Sampling plan and study population

Sensitivity and positive predicted value are commonly used to summarize the validity of registered causes of death.19,21 We therefore estimated the sample size based on expected values for these variables. Since there was no prior information on these measures in the Islamic Republic of Iran, we based our sample size calculations on expected sensitivity values of 50%, which suggested that about 200 deaths from a particular cause were required to adequately assess sensitivity at this level (95% confidence interval, CI: 42.5–57.5), or a sample size of 1400 deaths for the seven causes of interest. We initially selected 1800 cases into the study, in the expectation of some losses (25–30%) based on pilot study findings. We chose to conduct the study in district and provincial hospitals located in 10 out of 30 provinces in two regions of the Islamic Republic of Iran, based on willingness to participate in the research. Region 1 includes provinces in the north and north-west (East Azerbaijan, West Azerbaijan, Kurdistan and Zanjan provinces) and Region 2 includes provinces in the south, south-west and central parts (Bushehr, Hormozgan, Kuzestan, Kermanshah, Isfahan and Yazd provinces). We did not include eastern provinces owing to logistic considerations and the low probability of obtaining adequate medical records from hospitals located there. Hospitals were chosen on the basis of cases selected in chronological order from March 2005. This resulted in a total of 117 hospitals being selected from the 10 provinces.

Table 2 shows the distribution of study sample deaths from each region. Interestingly, the relative frequency for some causes varies considerably across the two regions. Thus, almost 80% of cases coded to “other and unspecified disorders of the circulatory system” were from Region 1. On the other hand, deaths in the sample coded to “heart failure” and “other cardiac diseases” appear to be much more common in Region 2. This suggests that physicians in different parts of the country assign different vague diagnoses for cardiovascular disease deaths. Overall, the study sample includes adequate numbers from each of these selected categories.

Data collection and processing

The reference period for the study was March 2005 to March 2006. The process of data collection and the design of the study are illustrated in Fig. 1. First, target cases for each province out of the total of 1800 deaths were calculated proportionally according to cause, based on data from the Iranian MOH&ME’s Death Registration System for 2005. For each selected death, data from this system, including date of death, hospital and registered underlying cause of death, were entered into the study database. From this database, personal identification details were used to trace medical records in health facilities.

Fig. 1. Study design for assessing misclassification of the seven ill-defined causes of death, the Islamic Republic of Iran, 2005-2006
ACME, Automatic Classification of Medical Entry; MOH&ME, Ministry of Health and Medical Education.

For inclusion into the study, the medical records of the deceased needed to contain a fully reported clinical history, medical examination report and/or one of the following clinical aids for diagnosis: electrocardiograph, radiology, computerized tomography (CT) scan or laboratory reports. A team of 15 physicians drawn from the 10 provinces was specifically trained to review and certify each death from the hospital records. A full-day workshop about the significance, aims and objectives of the research was held for reviewers (physicians) and other provincial staff involved in the study. Standard forms for collection of data and guidelines for reviewing medical records were prepared in Farsi and described in detail during the training workshop. Training was also provided in the principles and procedures for completing the international medical certificate of causes of death. During the training workshop, an assessment of inter-rater reliability was conducted to ensure consistency in death certification.

The medical records for each case were reviewed by one of the physicians on the study team (blind to the cause of death recorded at registration), who then specified the sequence of conditions on the death certificate.22 Reviewers also assessed the quality of clinical and paraclinical evidence from the medical records, which was categorized as: clinical history; electrocardiograph; X-ray and/or imaging; blood test; pathology; or other (specified).

The data for each case (in Farsi) were entered into a customized computer questionnaire in Epi Info™ version 3.3.2 (Centers for Disease Control and Prevention, Atlanta, GA, United States of America).23 All medical terms and conditions collected from the medical record review were translated from Farsi to English by a physician. The underlying cause of death was assigned by using ACME (Automatic Classification of Medical Entry) software, based on ICD-10 rules.2426 In about one-fifth of cases, the cause of death needed to be coded manually. This was done by Ardeshir Khosravi following a formal training course.

Staff at the Australian National Centre for Classification in Health in Brisbane, a WHO Collaborating Centre for the International Classification of Diseases, manually recoded a random sample of 150 cases (10%), to assess the quality of coding in assigning the underlying cause of death. There was 96% agreement between the two sources concerning the underlying cause of death, suggesting that our procedures for coding causes of death were valid and did not introduce serious bias.

Ethical approval for the study was obtained from the School of Population Health Research Ethics Committee (University of Queensland, Australia). The study was carried out with the full cooperation and support of the Iranian MOH&ME and the 11 Universities of Medical Sciences in the selected provinces.

Data analysis

Underlying causes of death from the two sources (death registration and medical records review) were aggregated according to the Iranian MOH&ME ICD Mortality Tabulation List.15 The few remaining deaths that were still classified to “senility without mention of psychosis” and “unknown” were aggregated into one group. Patterns of misclassification of causes of death were analysed for two broad age groups: 15–69 years, and 70 years and over. The causes of death selected for the study are relatively uncommon at younger ages, where mortality in the Islamic Republic of Iran is already comparatively low.27

Results

Adequate medical records were not available for 374 deaths in the target sample (21%), of which 147 cases were at ages 15–69 years. Of the remaining 1426 deaths, 655 (46%) were females and 771 (54%) were males. In the final study sample, only 106 (7%) deaths were at ages 0–14 years. The remaining deaths were roughly evenly divided across the two age groups: 15–69 years (582 deaths; 41%) and 70 years and over (738 deaths; 52%).

Table 3 shows the distribution of deaths reassigned to specific causes according to the strength of evidence contained in the medical records, summarized in three broad categories: confirmatory, suggestive or weak. Confirmatory evidence was available for one-quarter of cases of neoplasms; most cases of diabetes (86%), ischaemic heart disease (78%) and cerebrovascular disease (71%) were assigned on the basis of suggestive evidence. About 50% of the diagnoses of chronic obstructive pulmonary disease (COPD) and neoplasms were based on weak evidence from medical records.

Misclassification patterns by age groups

Accurate cause of death data for young and middle-aged adults is particularly important if public health policies and programmes to reduce premature mortality are to be appropriately informed. As Table 4 indicates, the probable underlying pattern of causes of death at these ages is considerably different to what the routine death registration system suggests. Of the 582 cases of ill-defined or vague diagnoses at these ages, less than 12% (69) were still classified as such after medical records review. The remainder were primarily reassigned to ischaemic heart disease (194, or 33%), cerebrovascular disease (75, or 13%) and, interestingly, injuries (56, or 10%). This undercoding of injury deaths in vital statistics has been observed in other countries of the region.2830 Half of these injury deaths had been assigned to the category “senility and unknown” in the death registration system. Other causes of death that were commonly misdiagnosed include diabetes, neoplasms and genitourinary diseases. While there is evidence of serious undercount of specific vascular disease deaths in registration data, the statistics for the broad category of cardiovascular diseases seem reliable: of the 388 cases of ill-defined vascular disease in the study sample, 345 or 90% were reassigned within the cardiovascular disease category upon medical records review.

A similar pattern of misclassification is apparent for deaths at older ages (70 years and over) as well (Table 5). Almost half of the 738 cases of ill-defined deaths were reassigned to ischaemic heart disease (219) and stroke (121), upon review. COPD (51) and digestive diseases (37) were also commonly misdiagnosed at these ages. Neoplasms and injuries, on the other hand, were less commonly misdiagnosed than at ages 15–69 years. Interestingly, the number of deaths from all ill-defined forms of cardiovascular disease in the routine registration system at these ages (460) was almost identical to the number of deaths reassigned to some form of cardiovascular disease (464) upon medical records review, confirming that mortality rates from the broad category of cardiovascular disease may not be that unreliable in registration data.

Discussion

Assessments of quality and reliability of data from routine death registration are critical.13 In the new Iranian MOH&ME’s Death Registration System (Deputy of Health), data on causes of death are collected from various sources and have been assessed to be about 80% complete.27 This study estimates misclassification patterns of seven ill-defined or vague diagnoses that were among the leading causes of death over the period 2000–2004.

Medical records review suggests that a substantial proportion (56%) of these vague and/or ill-defined causes of death can be reclassified to diseases of the circulatory system, primarily ischaemic heart disease (29%) and cerebrovascular disease (14%). These findings suggest that ischaemic heart disease probably causes about one-third more deaths, and stroke about 15% more, than recorded at routine death registration. Several other causes, including diabetes, injuries and COPD, are also likely to cause substantially more mortality than the official data suggest. These findings have important implications for epidemiological assessments and health planning in the Islamic Republic of Iran.

It is important to note that these correction factors are based solely on the redistribution of the seven selected ill-defined causes of death for this study. Death rates for leading causes of death may be even higher if there were to be systematic miscoding across specific well-defined diseases and injuries, as observed in other studies.8

Although traffic accidents are the third leading cause of death in registration data,15,31,32 our results suggest that mortality from external causes may be even higher than reported. A possible explanation might be the tendency to code injury deaths of undetermined intent to “unknown causes”, pending investigation by the Iranian Legal Medicine Organization, without subsequent correction based upon investigation results.

The hypothesis that deaths coded to “heart failure” and “hypertension” actually are due to more specific forms of vascular disease, particularly ischaemic heart disease and stroke, is confirmed by our investigation. Two-thirds of these vague diagnoses were reassigned to more specific causes within the cardiovascular disease category.

Our findings suggest a substantial undercount of major vascular diseases in the Islamic Republic of Iran, confirming findings from a smaller study carried out in Isfahan.33 In this study, a panel of physicians from the Isfahan Cardiovascular Research Center reviewed medical records of 571 deaths. They reported a positive predicted value of 0.82 for cardiovascular diseases, confirming substantially higher death rates from ischaemic heart disease and stroke than reported from the official records. According to the authors, the main reasons for this undercount were lack of experience in death certification and coding of the underlying causes of death.33

The Iranian death certificate15 has only one part with three lines for recording the sequence of events and conditions leading to death. This differs from the WHO standard death certificate,34 which provides space (Part II) for recording contributory causes of death. The absence of Part II in the Iranian death certificate results in contributory causes being mentioned in Part I of the certificate, which could affect the selection of the underlying cause of death. The MOH&ME would be well advised to adopt the international death certificate to reduce this potential source of error.

Limitations of the study

The validity of our study findings could be affected by the quality of documentation in medical records, which are subject to information bias since more than one person might collect information in the records and also different definitions might be used.35 Also, variation in the quality of medical records among hospitals in the different provinces could affect our findings.

The generalizability of our results to the entire population of the Islamic Republic of Iran depends on several factors, the most important being the selection of study deaths from hospitals. While nearly half of all deaths occur in hospitals, differences between the cause distribution of hospital deaths and deaths that occurred at home are likely; e.g. deaths that occur in hospital tend to be due to multiple end-stage conditions. Also, differential access to health-facility and in-patient services might result in disparities between hospital and non-hospital deaths. Generalizability might also be affected by the choice of provinces, which were selected on the basis of interest in the study, costs and the availability of other resources.

Our study has provided an audit of the quality of medical records in health facilities in the Islamic Republic of Iran. About one-quarter of records in the original sample were judged to be unusable, and this suggests an urgent need for training and procedural changes to improve hospital information systems. However, about half of these cases were deaths among elderly people who were referred to emergency units without complete medical records. This is an issue for medical records development in all countries, not only in the Islamic Republic of Iran. Nevertheless, 44% of rejected cases were deaths in young and middle-aged adults for which better health records ought to be available.

In this study, we used ACME software to assign the underlying cause of death. ACME is considered to be the de facto international standard for assigning causes of death.25,26 However, the software has some limitations.25 For some sequences, the causal relationship cannot be determined unequivocally from the ICD decision tables that are used for selecting the underlying cause of death, requiring the cause to be assigned manually. Additionally, the software strictly adheres to the selection rules and consequently there might well be overcoding of deaths based on “physiological derangement” at the expense of causes of death that are “etiologically specific”.25

We conclude that a significant proportion of deaths in the Islamic Republic of Iran classified to vague causes can be reclassified to more specific causes of death, primarily ischaemic heart diseases and stroke. There is an urgent need to improve the quality of medical records and cause of death certification. As a minimum, the standard international certificate of causes of death should be adopted, with appropriate training of physicians in its use. Until these changes are implemented, considerable caution should be exercised when using death registration data in the Islamic Republic of Iran for epidemiological research and planning. ■


Acknowledgements

We thank the staff of the 11 Universities of Medical Sciences (Ahvaz, Bushehr, Hormoszgan, Isfahan, Kashan, Kermanshah, Kurdistan, Tabriz, Urmia, Yazd and Zanjan) who participated in data collection and Amir Massoud Azad at the Islamic Republic of Iran MOH&ME who processed the data. We are especially grateful to Gary Waller from the Australian National Centre for Classification in Health in Brisbane who independently reviewed the quality of coding of the cause of death.

Funding: The authors gratefully acknowledge the Iranian MOH&ME for providing financial support for the study and also for providing a scholarship to support Ardeshir Khosravi’s doctoral studies in Australia.

Competing interests: None declared.

References

Affiliations

  • Iranian Ministry of Health and Medical Education, Tehran, the Islamic Republic of Iran.
  • School of Population Health, University of Queensland, Brisbane, Qld., Australia.
Share