Association of Exposure to Civil Conflict With Maternal Resilience and Maternal and Child Health and Health System Performance in Afghanistan

This survey study evaluates associations of conflict severity with improvement of health system performance, use of health services, and child nutrition outcomes in Afghanistan during the 2003 to 2018 reconstruction period.


Introduction
Humanitarian crises, including war, natural disasters, and population displacement, negatively affect population health and development. [1][2][3][4][5] Recent analyses have shown significant excess of child mortality in Africa due to conflict, 6 collapse of health systems, and deterioration of protective infrastructure that further exacerbate mortality and morbidity in conflict settings. [1][2][3] In the wake of increasing conflict and global instability, the need to understand how to make health systems more resilient and how to effectively shift from acute emergency response to systems building in chronic conflict zones is greater. One approach is to shift service delivery from centralized ministerial bodies to local nongovernmental organizations (NGOs), as in Lebanon 7 and Afghanistan 8-10 (eMethods 1 in the Supplement).
Afghanistan is a landlocked, impoverished South Central Asian nation of approximately 30 million civilians. For much of the last 4 decades, it has experienced conflict, insurgency, and war.
Nonetheless, the resilient nation has made notable gains in developing and rebuilding infrastructure during the past 15 years and has concurrently improved access to and quality of health care service delivery for women and children. [11][12][13][14][15][16] However, outreach and access challenges remain and are compounded by protracted conflict and its consequences in many provinces. 17 Several studies of population and health systems resilience in the wake of ongoing conflict in Afghanistan have been performed. Although cross-sectional and analytical studies exist, 14,15,[18][19][20][21][22][23][24][25][26][27][28][29][30] they have had limited scope, and few assessed the continuum of reproductive, maternal, and child health. We undertook a longitudinal assessment of Afghanistan's entire redevelopment period (2003-2018) using individual-level data sets from robust national survey data to further explore these gaps. We specifically assessed associations between ongoing conflict in Afghanistan and the levels and progress in maternal and child health service coverage, performance of health systems, and child nutrition during the 2003 to 2018 reconstruction period.

Study Design and Approach
This survey study used sequential panel data sets from large representative surveys to examine progress and trends during 2 critical transitionary periods in Afghanistan: the 2003 to 2010 immediate post-Taliban period when the bulk of reconstruction and development took place, and the 2010 to 2018 period, when the health financing models and security context changed. The association between changing levels of conflict and outcomes was examined at the ecological level (across study periods), at the macro level (across provinces and districts), and at the micro level (across households and health facilities). The protocol for the study was approved by the Aga Khan University ethics review committee, and waiver was obtained for the quantitative analysis of publicly available data sets. All participants provided written or oral informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Key Data Sources
We vetted all good-quality representative population-based health, nutrition, and health systems surveys conducted in Afghanistan after 2001. Quality assessment of relevant maternal and child health surveys in Afghanistan was conducted in an earlier study. 12 Surveys that aligned with our study periods, had adequate population coverage, and used standardized methods were considered in our analyses. For evaluation of reproductive, maternal, newborn, and child health service coverage, we Child nutrition data were obtained from the Afghanistan 2013 National Nutrition Survey. 34 The National Nutrition Survey was a large-scale nationally and provincially representative survey that collected population data on anthropometry, infant and young child feeding, and micronutrient deficiencies. The original individual or household was analyzed for all surveys, and available sample sizes are detailed in eTable 1 in the Supplement. For the Multiple Indicator Cluster Surveys and Afghanistan Health Survey, the purpose and content of the interviews were outlined to every participant, and informed consent was sought before any data collection. Indicators of health system performance were analyzed from the Afghanistan Balanced Scorecards (BSC) data sets, a series of annual (2004-2016) performance assessments of 6 key health service domains, consisting of 23 indicators and 3 summary indicators. 35 Each indicator is scored from 100, with higher scores indicating better performance. The sampling frame for the BSC consists of a random sample of patients and health workers from as many as 25 randomly selected comprehensive health centers, basic health centers, and health subcenters in each of the 34 provinces. Data are collected by surveyors from Kabul in more secure provinces and by trained local school teachers in less secure provinces.
Although BSC indicator definitions changed over time, generally, the 2004 to 2010 definitions are comparable, and so were the 2011 to 2016 definitions. For 2004 to 2016 annual survey rounds, data were available on 617 to 783 health facilities, 5719 to 7979 patient-health care professional interactions, 5597 to 7979 patient exit interviews, and 1452 to 2520 health worker interviews. 36 We analyzed all original BSC data and reconstructed indicators of health facility performance. An overview of all data sources used in this analysis is provided in eTable 2 in the Supplement.
We explored alternative classifications of conflict using measures of perception regarding fear collected as part of the 2018 Survey of the Afghan People 38 and the United Nations reports on security in Afghanistan 39 but ultimately decided that BRDs were a more objective measure. Further detail on our conflict classification is included in eMethods 2 in the Supplement.

Outcome Variables
Reproductive, maternal, newborn, and child health interventions coverage was examined using 10 standard indictors that span the continuum of care: contraceptive method (any or modern); antenatal care (ANC) by a skilled health care professional; facility delivery; skilled birth attendance (SBA); bacille Calmette-Guérin vaccination (BCG); diphtheria, pertussis, and tetanus vaccination (DPT3); measles vaccination; care-seeking for acute respiratory infection; oral rehydration therapy for diarrhea; and the Composite Coverage Index (CCI). The CCI is a commonly used composite of overall health coverage that includes curative and preventative child and maternal health interventions. 40 Domains and component indicators are defined in eTable 3 in the Supplement.
Performance of the health system was analyzed using standard BSC composite domains (client and community, human resources, physical capacity, quality of service provision, management systems, and overall mission) 35 and additional key component indicators as deemed relevant to health facility access and capacity in Afghanistan. We examined absolute mean differences for 6 broad health systems domains and an overall health system performance composite measure.

Domains and component indicators are defined in eTable 4 in the Supplement.
Child anthropometry indicators were examined as nutritional outcomes. We calculated z scores for height for age, weight for age, and weight for height according to World Health Organization child growth standards 41 for children younger than 5 years. Children more than 2 SD below the median of the World Health Organization reference population for height for age were categorized as stunted; more than 2 SD below the median of the reference population for weight for age, as underweight; or more than 2 SD below the median of the reference population for weight for height, as wasted. We also calculated the joint probability of stunting and wasting to examine concurrent malnutrition among Afghan children.

Statistical Analysis
To examine the association between conflict and child nutrition outcomes, we conducted a series of stepwise multivariable linear regression models using district-level (399 districts) data. Details are provided in eMethods 2 in the Supplement. [42][43][44] Univariate statistics were estimated using means (SDs) and frequencies or proportions as appropriate. We calculated mean differences, differences of mean differences, and annual percentage point changes in outcomes. We used 2-sided t tests and 1-way analysis of variance methods with ad hoc Tukey comparisons (type I error rate constrained at .05) to examine statistical differences across conflict subgroups.
To understand the association of conflict with the use of health services and health system performance outcomes, the cross-sectional surveys were assembled into panel data sets and the difference-in-differences (DID) analysis methods were used. 45 We withheld the assumption of parallel trends for all analyses (ie, absence of conflict would result in the mean outcomes of conflict and nonconflict provinces to follow parallel paths over time). In addition, we used interaction estimators in unadjusted and covariate-adjusted regression methods to estimate the DID effect. The general model specification included an interaction term between study period (categorical or continuous) and conflict dummy variables (moderate or severe intensity). Control variables were included as fixed or time-variant confounders measured at the facility or the province level.
Generalized linear models and generalized estimating equations were fitted through xtreg and xtgee routines, respectively, in Stata software, version 14 (StataCorp LLC). The reproductive, maternal, newborn, and child health intervention outcome models were adjusted for percentage of female illiteracy, rural population, and the type of NGO-contracting mechanism operating in the province (contracting in or out). The health system performance models were adjusted for patient volume, facility type (basic, comprehensive, or subcenter), NGO-contracting mechanism, and geographic region. Although a range of potential confounders were identified from the literature and expert opinion (eg, poverty, mean distance to city center, and distance to Kabul city), the above variables were included for parsimony and to avoid overfitting models. Effect estimates were reported with 95% CIs, and statistical significance was held at 2-sided P < .05 and P < .10 for borderline significance.

JAMA Network Open | Global Health
All analyses considered the survey design and weighting characteristics except pooled analyses, because design and weighting parameters differed between surveys. Statistical analyses were conducted from January 1 through April 30, 2019, using Stata software, version 14 (StataCorp LLC).

Results
Responses from a total of 64 815 women across surveys (mean

Use of Health Services and Conflict
The CCI improved in all but 2 provinces in both study periods, although coverage levels remained      Table 2). Crude box plots, adjusted means, and complete regression models are included in eFigure 5 and eTable 10 in the Supplement.   relative importance, conflict ranked about as important as maternal illiteracy (standardized βs, 0.09 and 0.11, respectively) as a correlate of wasting prevalence, and household poverty and lack of improved sanitation generally had very strong contributions to wasting (standardized βs, 0.30 and 0.29, respectively). No other child nutrition outcome was significantly associated with conflict. Of note, child stunting had an inverse association with conflict (ie, more conflict was associated with decreased stunting), which was statistically significant (β = −0.71 [95% CI, −1.38 to −0.03]) in the fully adjusted model.

Discussion
Several important findings emerge from our analysis. First, despite widespread conflict, Afghanistan has made significant progress in addressing key population needs for maternal and child health. regions. 54 The pathways linking increased conflict and insecurity to poor health facility infrastructure, client evaluations, and knowledge of health care professionals that were identified in our study require further investigation. Our findings that child emaciation increased with greater insecurity are similar to the findings of studies in other unstable and emergency contexts, [55][56][57][58] including assessments of recent conflicts in Somalia 55 and Nigeria, 56 and suggest that acute food deprivation is a continued challenge that requires immediate attention in conflicted provinces of Afghanistan.

Implications for Afghanistan
Afghanistan has had a successful model of contracting out health services to NGOs (eMethods 1 in the Supplement), which has effectively improved access despite security issues. 11,14 It appears that improving on and sustaining this model, particularly with the inclusive role of community health workers, will be a critical strategy to target the facility-and community-based health care interventions that are at risk of regressing in areas with severe-intensity conflict. 13 The community health workforce has played a critical role in overcoming health worker shortages in low-to middleincome countries, particularly in hard-to-access and insecure areas, and has been linked to notable improvements in maternal and child health in South Asia. 59,60 Strengthening the capacity of existing health facilities and quality of care is essential to further gains. Particular focus should be given to ensuring functional and sound facility infrastructure, appropriate assessment of patients, and improving the knowledge of health care professionals.
Availability and accessibility can be improved by building on the existing health service delivery mechanisms in the country. Health care staff, such as physicians, nurses, vaccinators, and technicians, have also fallen behind population needs, which may be associated with the observed reversal of gains found in our study; improvements should be urgently prioritized by the government. 18 Across all cadres, culturally appropriate efforts to encourage and attract support for female community health workers in remote public health settings are needed. Particularly in conflict areas, such initiatives could encourage the use and acceptability of family planning interventions.
Emergency medicine is one area of health care that is not considered in the basic package of health services and national health policies in Afghanistan 9,27 but should be included, given increasing conflict. Government and development partners must focus on peace and reconciliation efforts.
Mobilizing community Shuras (consultative councils) to mediate among government, NGOs, and armed oppositions groups has been tentatively successful in Afghanistan, although perhaps not sustainable. 24

Strengths and Limitations
Several limitations should be recognized in considering these findings. Given the lack of a reliable district health management information system and a formal mechanism for civic registration and vital statistics, 61 we mainly relied on available household survey data. The BSC provided standardized annual data on health facilities; however, no linkage of facilities to household data in catchment populations was available, thus limiting inferences about population. Despite this, our unique pseudocohort approach to linking several good-quality, large-scale surveys on health and the health system in Afghanistan is a major strength of this study; it permitted otherwise diverse crosssectional surveys to be analyzed concurrently with a narrative of a comprehensive story of change across decades.
We relied on Uppsala's BRDs as a means of classifying the severity of conflict, but we recognize this data set's limited ability to represent the full dimensions of conflict and insecurity. Nonetheless, we examined several classifications of conflict severity using BRDs in sensitivity analyses with consistent findings, which strengthens our results.
Future data collection efforts should attempt to collect and triangulate household data to assess Our multivariable quantitative analysis used the increasingly common and powerful DID methods to control for baseline values and secular trends in outcomes when assessing the association of an environmental shock (ie, conflict). 62 We also have large sample sizes from panel cross-sectional surveys, which strengthened statistical power and inferences. Nonetheless, a few limitations should be noted. The parallel trends assumption was difficult to assess because pre-2003 data in Afghanistan were virtually nonexistent. The provinces in each conflict severity group were also different in the 2 study periods, given that conflict was dynamic and varied across time.
Notwithstanding these limitations, we believe that because the main objective of this study was to assess associations with conflict severity, staying true to conflict criteria was more important than retaining consistent provinces.

Conclusions
Afghanistan has made progress despite ongoing insecurity and conflict, yet several essential health interventions and the health system remain at risk of falling behind. Immediate attention and targeting appear to be needed to maintain and scale up gains toward achieving the United Nations' Sustainable Development Goals. This will undoubtedly necessitate effective and strategic collaboration from international bodies, funders, government, NGOs, civilians, and local leaders.