Potentially Avertable Child Mortality Associated with Surgical Workforce Scale-up in Low- and Middle-Income Countries: A Global Study

Abstract

Background

Expansion of access to surgical care can improve health outcomes, although the impact that scale-up of the surgical workforce will have on child mortality is poorly defined. In this study, we estimate the number of child deaths potentially avertable by increasing the surgical workforce globally to meet targets proposed by the Lancet Commission on Global Surgery.

Methods

To estimate the number of deaths potentially avertable through increases in the surgical workforce, we used log-linear regression to model the association between surgeon, anesthetist and obstetrician workforce (SAO) density and surgically amenable under-5 mortality rate (U5MR), infant mortality rate (IMR), and neonatal mortality rate (NMR) for 192 countries adjusting for potential confounders of childhood mortality, including the non-surgical workforce (physicians, nurses/midwives, community health workers), gross national income per capita, poverty rate, female literacy rate, health expenditure per capita, percentage of urban population, number of surgical operations, and hospital bed density. Surgically amenable mortality was determined using mortality estimates from the UN Inter-agency Group for Child Mortality Estimation adjusted by the proportion of deaths in each country due to communicable causes unlikely to be amenable to surgical care. Estimates of mortality reduction due to upscaling surgical care to support the Lancet Commission on Global Surgery (LCoGS) minimum target of 20–40 SAO/100,000 were calculated accounting for potential increases in surgical volume associated with surgical workforce expansion.

Results

Increasing SAO workforce density was independently associated with lower surgically amenable U5MR as well as NMR (p < 0.01 for each model). When accounting for concomitant increases in surgical volume, scale-up of the surgical workforce to 20–40 SAO/100,000 could potentially prevent between 262,709 (95% CI 229,643–295,434) and 519,629 (465,046–573,919) under 5 deaths annually. The majority (61%) of deaths averted would be neonatal deaths.

Conclusion

Scale up of surgical workforce may substantially decrease childhood mortality rates around the world. Our analysis suggests that scale-up of surgical delivery through increase in the SAO workforce could prevent over 500,000 children from dying before the age of 5 annually. This would represent significant progress toward meeting global child mortality reduction targets.

Appendices

Appendix 1: Imputation Density Plots for Imputed Variables

Appendix 2: Methods for Estimation of Mortality

We used the following formula to calculate deaths averted where Y is the under-5 mortality rate (deaths for 1000 live births) for a given country, X the surgeon/anaesthesiologist/obstetrician (SAO) density, and Z the remaining covariates in the model, and where E denotes expected value (i.e., average).

$${\mathbb{E}}(\log Y|X,W,{\varvec{Z}}) = \alpha X + \gamma W + \beta^{T} {\varvec{Z}}$$

The parameter α represents the mean change in log Y associated with an increase of X by one unit (number of SAOs per 100,000 population), keeping the remaining variables Z fixed. Thus, for any baseline SAO density x, and potential increase r according to the formula:

$$(1 - e^{r\alpha + s\gamma } ){\mathbb{E}}(Y|X = x,W = w,{\varvec{Z}}) \approx {\mathbb{E}}(Y|X = x,W = w,{\varvec{Z}}) - {\mathbb{E}}(Y|X = x + r,W = w + s,{\varvec{Z}})$$

To estimate potentially avertable deaths across countries, every country that has SAO below 20 (or 40), we can take r = 20 − X (or 40 − X), and plug in the estimated α into the formula above, and the observed Y in place of the baseline E(Y | X = x, Z), to obtain an expected decrease in under-5 mortality rate associated with increasing SAO up to 20 (or 40).

To convert this to a raw number of annual deaths prevented by such an increase in SAO, we can replace Y with the raw number of observed annual deaths, Y*.

$$(1 - e^{{r\widehat{\alpha } + s\widehat{\gamma }}} )Y^{*} \approx Y^{*} - {\mathbb{E}}(Y^{*} |X = x + r,W = w + s,{\varvec{Z}})$$

Note that Y* = TY/1000, where T is the total number of live births in a year. Multiplying the above approximation by T/1000, we can use the following formula where the right hand side expresses the expected decrease in under-5 deaths associated with an increase in SAO density by r. The same method was used substituting for infant and neonatal mortality estimates. We adjusted our predictions for the potential change in procedures associated with scale up of surgical workforce in order to take into account non workforce related factors.

$${\mathbb{E}}[\log W|X] = a_{0} + a_{1} \log X$$

Appendix 3: Sensitivity Analysis

Model scenario at extremes of surgically amenable child mortality

Model	U5MR	Lower CI of mortality*	Upper CI of mortality*
Under 5 Mortality rsate	− 0.0045 (0.0027)	− 0.00437 (0.0037)	− 0.0048 (0.006)
Infant mortality Rate	− 0.0046 (0.0017)	− 0.0044 (0.002)	− 0.0047 (0.005)
Neonatal Mortality Rate	− 0.0048 (0.0076)	− 0.0046 (0.01)	− 0.005 (0.012)

1. *Sensitivities based on use of upper and lower confidence intervals compared to point estimates for mortality estimates by the UN Agency for Child Mortality. 2018 estimates were used for all calculations. Model beta coefficients and p values are reported

Delta adjustment technique In order to test if the imputed values can be considered to be missing at random conditional on the variables included in the imputation model, each value for SAO density was adjusted by 0, − 10, − 20 and − 50 and the mean imputed values across 5 imputations was compared. The mean SAO density on complete case analysis was 37.4 with a difference of − 1.77 among imputed values for a delta adjustment of 0 (MAR). There were no significant increases in the difference for deltas up to 20 SAO density, suggesting that missing SAO variables can be considered missing at random conditional on the variables included in the imputation model (“Appendix 3”).

Delta Adjustment technique
Realized difference in means of the observed and imputed SAO density data under various delta adjustments. Based on 5 imputations
Delta adjustment	Mean SAO density	Difference
0 (MAR)	35.27	− 1.77
− 10	34.922	− 2.118
− 20	33.04	− 4
− 50	29.885	− 7.155