Unintentional injury mortality in India, 2005: Nationally representative mortality survey of 1.1 million homes

Study setting and data collection

The Registrar General of India (RGI) randomly selected 6671 small areas from approximately one million small areas defined in the 1991 census for its Sample Registration System (SRS) [9]. In 1993, household characteristics of the SRS areas, each with about 150 houses and 1000 people, were documented. The SRS sample frame covered 6.3 million people in all 28 states and seven union territories of India. SRS sampling frame is based on the results of census of India, which is conducted every ten years. The selected households are continuously monitored for vital events by two independent surveyors. The first is a part-time enumerator (commonly a resident of the village/area or a local school teacher familiar with the area/village) who visits the households every month. The second is a full-time (nonmedical) Registrar General of India’s surveyor who visits the households every 6 months. Another staff member from the office of Registrar General of India does the reconciliation of vital events reported by the part-time enumerator and the full-time surveyor, arriving at a final list of births and deaths for each household, at the completion each half-yearly survey.

In the last decade, the RGI has introduced an advanced form of verbal autopsy called “RHIME” (Routine, Reliable, Representative and Re-sampled Household Investigation of Mortality with Medical Evaluation) [9, 10]. Verbal autopsy is a method of ascertaining the cause of death by seeking information on signs, symptoms and circumstances from a family member or care taker of the deceased [10]. Since 2001, about 800 non-medical graduates who were full time employees of the RGI, had knowledge of local languages and were trained to implement the RHIME method visited the families to record events preceding each death using three age specific questionnaires (neonatal, child and adult) including a narrative in the local language. The neonatal (0–28 days) and child death (29 days–14 years) questionnaires included a direct question, “Did s/he die from an injury or accident? If yes, what was the kind of injury or accident?” Response options included 1) Road traffic accident 2) Falls 3) Fall of objects (on to the person) 4) Burns 5) Drowning 6) Poisoning 7) Bite/sting 8) Natural disaster 9) Homicide/assault 10) Unknown. Place of death was recorded in all deaths with response options of 1) Home 2) Health facility (like government hospitals, private hospitals and registered practitioners) and 3) Others (including road side, public area, on transportation and body of water). A random sample of about 5% of the areas was resurveyed independently generally with consistent results. Details of the methods, validation results, and preliminary results for various diseases have been reported previously [10–14].

Cause of death assignment

The field reports including the individual narratives were sent randomly based on the language, to at least two of 130 physicians who were specially trained in disease coding. The physicians assessed the underlying cause of death and assigned a three character code from the International Classification of disease (ICD), 10^th Revision [15]. Unintentional injury deaths were allocated ICD codes from chapter XX for external causes of morbidity and mortality, including V01-X59, Y40-Y86, Y88, and Y89 codes. In case of chapter level disagreement between the two physicians, the final ICD code was assigned by a third senior physician. In case of sub chapter disagreements, specific codes were adjudicated by a specific codes were adjudicated two members of the research team.

Reports could not be collected for 12% of the identified deaths mostly due to migration of the household or change of residence; this is unlikely to have led to any systematic misclassification in cause of death as these missing deaths were distributed across all states.

Moreover, the SRS definition of usual resident included those who travel away from home for periodic work [9], so deaths away from home were captured provided the whole household had not moved. A total of about 136,480 deaths were identified between January 1, 2001 and December 31, 2003. About 9% of all the death reports could not be coded due to field problems such as poor image quality of the narrative or insufficient information; hence cause of death was identified for 122,828 deaths.

National estimates for absolute number of deaths and mortality rates

We applied the proportion of each cause of death to the UN Population Division estimates of Indian deaths from all causes in 2005 ((9.8 million; upper and lower limits 9.4 and 10.3 million respectively) [13, 16]. UN estimates were used for more accurate calculation of deaths and mortality rates (using Preston and Coale method) [17] because the SRS undercounts mortality by approximately 10% [18, 19]. All major cause of death like malaria, HIV and child mortality [12, 13] has been estimated for the year 2005 from the Million Death Study making cause specific mortality comparable for policy implications.

The proportion of cause specific deaths in each age category was weighted to the SRS sampling fractions in the rural and urban parts of each state. However, unweighted proportions yielded nearly identical results [20]. Application of the data from 2001–03 to 2005 deaths should not introduce major bias as there was little change in the yearly distribution of cause of death in the present study (p = 0.736 for yearly variation in proportional mortality). Sub national estimates were produced for six major regions (north, south, west, central, northeast, and east) [9] from segregating the national UN totals by the relative SRS death rates, as described earlier [21]. Live births totals from UN were used to calculate mortality rates for children younger than five years [21, 22]. Confidence interval (95%) for each cause of death proportion or mortality rate was calculated (using the Taylor linearization method) on the basis of the survey design and the observed sample deaths in the present study [23].

SRS enrolment is on a voluntary basis, and its confidentiality and consent procedures are defined as part of the Registration of Births and Deaths Act, 1969. Oral consent was obtained in the first SRS sample frame. Families are free to withdraw from the study, but the compliance is close to 100%. The study poses no or minimal risks to enrolled subjects. All personal identifiers present in the raw data are anonymised before analysis. The study has been approved by the review boards of the Post-Graduate Institute of Medical Education and Research, St. Michael’s Hospital and the Indian Council of Medical Research.

Limitations

Our study had some limitations. Verbal autopsy methods are known to misclassify some causes of death among neonates and older age groups of 70 years and above [10, 34]. Earlier comparisons of verbal autopsy to urban hospital records in India indicated a sensitivity of 85% and specificity of over 95% for injuries [35]. We caution, however, that hospital-based studies are not ideal studies as a large majority of deaths occur in India without medical consultation [36] and also because of the differences observed in age-sex composition of injury deaths recorded in health facility versus those recorded at home (Additional file 3: Table S1). Misclassification for injuries overall has been low in verbal autopsy validation studies elsewhere, with the exception of falls where some misclassification with cerebro-vascular conditions was reported [37]. We estimate the injury deaths for year 2005 using proportionate injury mortality recorded during 2001-03. We did not observe any change in proportionate mortality from 2001 to 2003, hence, assumed that proportionate mortality would not have changed in next two year also, however, that may not be the case.

The disease burden in India is undergoing a transition with the burden of both chronic conditions and injury rapidly rising. However, injury is a neglected epidemic in India and few resources are dedicated towards prevention or treatment of injuries. Our results provide reliable national and regional estimates of injury mortality which can inform the allocation of resources and development of an evidence based national and state policy for injury control. Our results suggest upward revision is needed of falls mortality and perhaps downward revision of fire related deaths. Future research should aim to formulate effective injury surveillance systems, epidemiological assessment of all outcomes of injuries, advocacy for prevention and treatment and appraisal of existing effective interventions for injury prevention and trauma care.