A review of national health surveys in India

Abstract Several rounds of national health surveys have generated a vast amount of data in India since 1992. We describe and compare the key health information gathered, assess the availability of health data in the public domain, and review publications resulting from the National Family Health Survey (NFHS), the District Level Household Survey (DLHS) and the Annual Health Survey (AHS). We highlight issues that need attention to improve the usefulness of the surveys in monitoring changing trends in India’s disease burden: (i) inadequate coverage of noncommunicable diseases, injuries and some major communicable diseases; (ii) modest comparability between surveys on the key themes of child and maternal mortality and immunization to understand trends over time; (iii) short time intervals between the most recent survey rounds; and (iv) delays in making individual-level data available for analysis in the public domain. We identified 337 publications using NFHS data, in contrast only 48 and three publications were using data from the DLHS and AHS respectively. As national surveys are resource-intensive, it would be prudent to maximize their benefits. We suggest that India plan for a single major national health survey at five-year intervals in consultation with key stakeholders. This could cover additional major causes of the disease burden and their risk factors, as well as causes of death and adult mortality rate estimation. If done in a standardized manner, such a survey would provide useable and timely data to inform health interventions and facilitate assessment of their impact on population health.


Introduction
Health information gathering is an important part of any health system, but is often weak in low-income countries, plagued by poor quality data that are inadequate for informing health policy. [1][2][3][4] Population-based surveys are an invaluable source of health information. 5 A key aim of these surveys is to provide high-quality data for policy development and programme planning, monitoring and evaluation. Population-based surveys have been used extensively to gather information on fertility, mortality, family planning, maternal and child health, and some other aspects of health, nutrition and health care in India. 6 We have previously reported that the health information system in India has not kept up with the epidemiological transition. 6 In this paper, we assess national health surveys conducted in India since 1992 that were designed to provide information on health indicators at subnational levels. We describe and compare the health information covered by these surveys over time, the availability of resulting data in the public domain and the use of these survey data in publications. Based on our findings, we highlight the issues that need consideration to improve the usefulness of these surveys. We believe they should be able to provide more effective, useable and timely data on the health status of the population, given the evolving disease burden in India.

Reviewing surveys
We selected large-scale, national, population-based household surveys that provided data on health indicators at the subnational levels in India from 1992 to 2015. These were the National Family Health Survey (NFHS), the District Level Household Survey (DLHS) and the Annual Health Survey (AHS), (Box 1). The surveys are summarized in Table 1.

Themes
We reviewed the survey questionnaires to assess: survey period and sample sizes; types of respondent; key themes; timeframe for availability of data in the public domain; and analytical publications resulting from the data. A more detailed review of the number of children, reference period and age groups was undertaken to gauge the utility of the data for assessment of trends in child mortality, maternal mortality and immunization.
To determine how well the household, male and female survey questionnaires corresponded to disease burden in the country, we assessed the proportion of questions covering major themes: maternal and child health; reproductive health other than infections; reproductive tract infections; other adult infections; noncommunicable diseases; and injuries. Data on anthropometric and biological markers were analysed in addition to the questionnaires.
We recorded the time between the completion of data collection for each survey round and the availability of individual-level data in the public domain. We conducted a PubMed database search to identify peer-reviewed research papers from January 1993 to March 2015 that had used data from either or all of the first three rounds of the NFHS and DLHS. For the AHS, this search was done for research papers published between January 2011 and March 2015. The fourth rounds of the NFHS and DLHS were not included in this search as the data collection for the former is not yet complete and the data for the latter have not yet been released in the public domain. The search terms used in PubMed to identify relevant publications were "National Family Health Survey" or "NFHS and India", "District Level Household Survey" or "DLHS and India" and "Annual Health Survey" or "AHS and India". We screened the titles and abstracts of identified articles and reviewed the full texts of those that analysed data from the surveys. Review papers and the papers that merely made reference to survey data in background or discussion sections were excluded.

Survey characteristics Survey period
The first survey conducted was NFHS-1 in 1992-1993. The following three rounds of NHFS were done in 6-9 year intervals, which were longer than the DLHS interval of 4-5 years. The period of the first DLHS survey overlapped with NFHS-2 and the following survey rounds done with close proximity. The AHS, which is complementary to DLHS-4, was initially done in 2010-11, with two further rounds between 2011 and 2013 ( Table 1).

Types of respondents
There were some changes in the types of respondents across these surveys over time (Table 2). Ever-married women were surveyed in all rounds of the NFHS and AHS. DLHS-1 and DLHS-2 surveyed only currently married women but DLHS-3 and DLHS-4 surveyed ever-married women. NFHS-3, NFHS-4 and DLHS-3 also included never-married women.
The ever and/or currently married women interviewed in all surveys were of reproductive age; however, the age boundaries for inclusion varied both across surveys and between different rounds of the same survey. Women up to 49 years of age were selected as respondents in all rounds of the NFHS; the lower age limit for NFHS-1 was 13 years, which was raised to 15 years during subsequent rounds. Women aged 15-44 years were surveyed during the first two rounds of the DLHS and the upper age limit was raised to 49 years for ever-married women in DLHS-3 and DLHS-4. The age group for never-married women was 15-24 years in DLHS-3. The AHS surveyed women 15-49 years of age.
Male representation across surveys has been inconsistent. Only in four rounds have men been represented. Men aged 15-54 years were interviewed in NFHS-3 and NFHS-4. Men aged 20-54 years were interviewed in DLHS-1, whereas the husbands of eligible women, regardless of age, were interviewed in DLHS-2. The AHS did not interview men.

Interview
The key survey themes are shown in Table 2 and Table 3. The numbers of disease-or condition-specific questions increased in all surveys over time. From 246 to 868 questions in the NFHS, from 200 to 339 questions in the DLHS and from 137 to 207 questions in the AHS. Of these questions, more than 90.5% of questions were about maternal and child health and reproductive health (range: 90.6-99.1%; Table 3). Adult infections other than those of the reproductive tract received very little attention in the surveys, constituting only 0.6-3.0% of the total disease-or condition-specific questions. The number of questions on noncommunicable diseases increased for each round in the NFHS and DLHS, from two to 41 and zero to 10, respectively.
NFHS-4, DLHS-4 and the AHS baseline had questions on tobacco and alcohol use, which are major risk factors for chronic diseases. However, these questions did not fully meet the criteria for the STEPwise approach to surveillance, recommended by the World Health Organization (WHO) for monitoring risk factors over time. 22,23 Only NFHS-3 and NFHS-4 had more than three questions related to injury (24 and

Box 1. Description of major surveys done in India between 1992 and 2016
The National Family Health Survey (NFHS) is the equivalent of demographic and health surveys done in many countries around the world. 7 The NFHS is overseen by the Ministry of Health and is coordinated by the International Institute for Population Sciences (IIPS) in Mumbai, as the nodal agency, with support from ORC Macro and other agencies. 8 The primary aim of the NFHS has been to provide information on maternal and child health and reproductive health. Three rounds of the NFHS were conducted in 1992-1993, 1998-1999 and 2005-2006, and the fourth round is currently underway. 9-12 The first three rounds of the NFHS were designed to provide state level data, but the fourth round, with a much larger sample size, will generate estimates of most indicators for all 640 districts in the country. 12 The District Level Household Survey (DLHS) was launched in response to the need for districtlevel data on the Reproductive and Child Health Programme. 13 The DLHS is carried out by the International Institute for Population Sciences with oversight by the Ministry of Health. Four rounds of DLHS have been undertaken: 1998-1999, 2002-2004, 2007-2008 and 2012-2014. [14][15][16][17] The fourth round was done in coordination with the Annual Health Survey (AHS), with the former not conducted in nine states covered by the latter. The AHS has been conducted in the less developed states of India (Assam, Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh, Odisha, Rajasthan, Uttar Pradesh and Uttarakhand). 18 The sample sizes at the district level in the AHS are much larger than those in the DLHS and aim to generate more robust estimates at the district level, especially of infant mortality.   (Table 3). Questions on antenatal care, delivery and postnatal care, birth history and family planning were included in all surveys with the exception of postnatal care in NFHS-1. Key subthemes regarding child health were immunization, breastfeeding practices and common childhood morbidity symptoms (cough, fever and diarrhoea).
All rounds of the NFHS included questions on women's employment status and fertility preferences. Rounds 2, 3 and 4 of the NFHS included questions on quality of health services, sexually transmitted infections (STIs), human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS) and women's autonomy. Several common themes were identified in the separate questionnaires completed by both women and men during NFHS-3 and NFHS-4: reproductive history; marital and sexual relationships; co-habitation; family planning and fertility preferences; quality of health services; STIs and HIV/ AIDS. Additional themes in the men's questionnaire were male involvement in health care and male attitudes towards women's autonomy and domestic violence (Table 2).
DLHS-1 and DLHS-2 included questions on the quality of public sector health services; however, these were dropped in subsequent rounds. Several new themes were added to DLHS-3, including sex education, age at marriage, infertility, obstetric fistula, knowledge about reproduction and public sector health programmes; these were all dropped in DLHS-4. Additional information on fertility preferences and menstruation was documented in DLHS-3 and DLHS-4. The husbands' questionnaire in DLHS-2 collected data on family planning and fertility preferences and on STIs and HIV/AIDS. In addition to the core themes of maternal and child health, birth registration was documented in the AHS.

Anthropometry and biomarkers
Height and weight were measured for children during all rounds of the NFHS, though the age varied in the different rounds (Table 4). Height and weight were measured for men and women in NFHS-2, NFHS-3 and NFHS-4. DLHS-1 and DLHS-3 did not include any anthropometric measurements.
DLHS-2 included weight measurement only for children younger than 6 years to calculate weight-for-age as an indicator of nutritional status. Height and weight were measured in DLHS-4 and in a subsample of households in the AHS for children 1 month of age and older, as well as women and men.
The surveys evaluated various biomarkers, especially in the later rounds (Table 4). NFHS-2 included assessment of anaemia among children 6-35 months of age and ever-married women 15-49 years of age. Anaemia testing was also done for men in NFHS-3 and NFHS-4. Anaemia testing was done for children, girls and women in DLHS-2, but not in DLHS-3. DLHS-4 and the AHS included anaemia testing for children 6 months or older as well as women and men. HIV testing was included in a subsample of men and women in NFHS-3 and NFHS-4. Blood pressure measurement and blood testing for fasting plasma glucose were done in men and women in NFHS-4 and DLHS-4 and in a subsample of men and women in the AHS.

Trend analyses Estimating child mortality
The information collected on deaths and age at death among all children born to ever-married women 15-49 years of age in their lifetime is consistent across all rounds of the NFHS, clarifying trends in child mortality over time using the lifetime data on births. In contrast, the data on birth histories varied in the different DLHS rounds, ranging from the preceding 3 years of the survey to lifetime data. The AHS baseline round collected birth history information for the preceding 3-year period, and the update rounds captured this information for the preceding year. On assessing the comparability of childhood mortality indicators across all rounds of the NFHS, DLHS and AHS, analogous estimates can be generated only for 3 years preceding the surveys for currently married women aged 15-44 years (Table 5).

Estimating maternal mortality
Comparable estimates of maternal deaths in the 2 years preceding the survey among women aged 15-49 years are possible using NFHS-1, NFHS-2 and NFHS-4, but maternal death data were not collected in NFHS-3. In the various rounds of the DLHS, the reference period for the collection of data on maternal deaths varies from 1 to 3 years preceding the survey. In the AHS, the reference period for maternal deaths ranges from 1 to 5 years preceding the last survey (Table 5).

Immunization
Assessment of immunization trends over time using all the NFHS, DLHS and AHS rounds is possible only for the last two surviving children born in the 3 years preceding the surveys, due to variation in the reference periods and in the number of births and living children for which immunization data were collected in the various rounds ( Table 5).

Timeliness of data availability
Individual-level NFHS and DLHS data -without individual identifiers to maintain participants' confidentialityhave to be made available in the public domain for analytical use.

Survey data publications
We identified 600, 95 and 73 publications resulting from the NFHS, DLHS and AHS respectively. Based on the review of the title and abstract, 337, 48 and three publications had used the NFHS, DLHS and AHS data, respectively; we reviewed the full text of these publications. Data from only NFHS-1 were used in 56 articles, data from NFHS-2 in 83 articles and data from NFHS-3 in 145 articles. The remaining 53 publications used data from two or more of the NFHS rounds. Only data from DLHS-2 and/or DLHS-3 were used in publications. No publication using DLHS-1 data was identified. One publication used AHS baseline survey data and two used the first update of the AHS survey data.

Policy & practice
National health surveys in India Rakhi Dandona et al.

Discussion
The national population-based health surveys in India started a quarter of a century ago with a predominant focus on maternal and child health, as these were considered the most visible and prominent health problems at that time. Over this period, the disease burden has shifted significantly towards noncommunicable diseases. Data from the global burden of disease study suggest that in India in 1990, diseases among children younger than 15 years and maternal disorders accounted for 57% of the total disease burden (with about 60% of this in the first year of life). In 2013, this burden had decreased to 33% of total disease burden, while noncommunicable diseases made up 52% of the total disease burden. 24 However, in the latest national health surveys, questions on noncommunicable diseases constituted less than 5% of the total questions. Similarly, injuries are barely represented in national health surveys even though these contributed 13% of the total disease burden in 2013. 24 While estimation of disease burden should not be the only criterion for inclusion in large-scale national health surveys, not having nationwide estimates for the conditions causing major disease burden is problematic. Reliable nationwide population-based data on major noncommunicable diseases, such as ischaemic heart disease, chronic obstructive pulmonary disease, stroke, low-back and neck pain and depression are scanty in India, as are similar data on injuries. Such data are also unavailable for tuberculosis and pneumonia. 25 Attempts to improve coverage of noncommunicable diseases in national health surveys are a move in the right direction, but more could be done. The surveys could be expanded to meet WHO's criteria for monitoring of noncommunicable diseases, the STEPwise approach to surveillance. This approach includes standardized data on four behavioural risk factors (tobacco use, alcohol use, low fruit and vegetable intake and physical inactivity) and four biological risk factors (body mass index, blood pressure, fasting blood glucose and blood cholesterol). 22 Among the behavioural risk factors, tobacco and alcohol use are being assessed in national health surveys, but these do not fully meet the STEPS standardized data criteria. Low fruit and vegetable intake and physical inactivity are not yet being measured. Among the biological risk factors besides body mass index (which has been included in most surveys), blood pressure and fasting blood glucose have been added in the most recent rounds of the national surveys, but blood cholesterol is still not included. Recent national health surveys have only partly addressed these data gaps since our previous report, which preceded these surveys. 23 National health surveys have the potential to increase data on disease burden by including biomarker measurements and diagnostic tests. For example, inclusion of HIV testing in NFHS-3 enabled a more accurate estimation of HIV prevalence. 26,27 Rapid diagnostic tests for tuberculosis 25,28 and malaria 29 and assays for measuring blood lipids in the field 30 could also be included. While a detailed assessment of all major diseases is not feasible in a single national survey, opportunities exist for adding additional categories of information. Some countries use a range of clinical and biomarker tests in their surveys and some regularly rotate the health and/or disease topics between rounds to make each round more manageable and frequent. [31][32][33][34][35][36][37][38][39] Reliable cause-of-death data are important for informing decisionmakers. India lacks an effective vital registration system that can provide such data across the country. To increase data on cause of death, automated algorithms could be used, which enable researchers to assign cause of death from large-scale verbal autopsy data. This is both more reliable and efficient than the resource-intensive physician-coding methods. 40,41 Comparability of measurements over time and across population groups is fundamental to optimal interpretation and use of survey data. 2,42,43 Given the enormous amount of data collected in national surveys, we calculated the feasibility of trends assessment over time for child mortality, maternal mortality and immunization between and within these surveys. All rounds of the NFHS had documented birth history consistently, allowing for comparable estimates over time as all births were captured with no restriction on reference period. However, the DLHS and AHS rounds captured births and/or pregnancies only for specific reference periods, which varied within and between surveys, thereby limiting the potential for using all the collected data for this purpose. Similarly, the reference period for data on immunization coverage varied within and between the surveys.
A systematic review reports that among publications in PubMed concerning global demographic and health surveys, there were many using the Indian NFHS data. 44 We report 336 original research publications using NFHS data. On the other hand, the three DLHS rounds completed to date resulted in only 48 publications. This is puzzling, given that DLHS surveys were designed to provide district-level estimates, whereas the first three NFHS rounds, with smaller sample sizes, were designed only to provide state-level estimates. One of the reasons for the poorer use of DLHS could be that the data are made available in a format which is not user friendly. The AHS has provided individual-level data in the public domain only very recently, so the low number of publications from these data is not surprising.
The time between completion of data collection and individual-level data availability for analysis by researchers decreased between the first and third rounds of the NFHS and DLHS, but increased again for the last round of the DLHS. Part of the reason for this delay could be the effort needed to synchronize the DLHS-4 data with the AHS data, as these two surveys are complementary, with each covering approximately half the country's population. In any case, such delays in use of a public good resource should be avoided. The recent availability of individual-level AHS data on request is a positive step towards increasing the effectiveness of the data.

Conclusion
As national health surveys are resourceintensive, it would be wise to maximize the knowledge gained from them that could be used to improve population health in India. We propose that consultation -similar to the consultative development process underpinning the National Health Survey in Brazil 45 -could improve the design of national health surveys in India. We have several recommendations. First, instead of having multiple, frequent surveys with overlapping goals, India should have a single major national health survey at five-year intervals. This could provide data on additional major causes of disease burden and their risk factors, along with cause-of-death data using automated verbal autopsy methods and include adult mortality rate estimation. The sample sizes should aim to provide state-level estimates for all indicators and district-level estimates for crucial indicators to capture the key features of health status heterogeneity across the country. Second, data collection on the key variables should be standardized to meet monitoring standards and to provide comparable data over time. Third, effective partnerships with a larger range of relevant stakeholders, including the academic community, should be established to increase the relevance and usefulness of the data. Fourth, detailed methods should be published. Fifth, individuallevel data from these surveys should be made publicly available as soon as possible so that it can be used in the urgent tasks of informing policy and developing a more effective health system. Sixth, linking household survey data with health service use and administrative data, preferably using geospatial coding methods could be considered. Over time, India could also consider a continuous design for its national health survey, with advantages for survey management and timely provision of findings. 46 ■