Sergio Bautista-Arredondo INSP VMMC unit costs variation and determinants using facility-level, primary data from several studies. Sergio Bautista-Arredondo INSP
Motivation Objective Limitations of secondary data To estimate determinants of unit cost variation and predict cost curves for different scenarios Scale Other service provision characteristics Limitations of secondary data Mean aggregated data not granular enough for analysis Country variation but not enough variation (or reporting) of other characteristics No single study could address this objective
Data Obtainment and Standardization
Data Obtainment and Standardization
Data Obtainment and Standardization
Description of data Observations 220 Urbanicity Rural Facilities 50% Observations 220 Urbanicity Rural Facilities 50% Urban Facilities Ownership Private facilities 37% Public facilities 63% Facility type Hospitals 47% Clinics 53% Mean Unit cost per VMMC in USD (mean) 66 Average VMMC per year 1097 Number of studies 7 Countries 8 Description of data
Primary data
Determinans of unit cost We conducted OLS and GLM regression models to estimate the association between facilities’ characteristics and unit cost variation across sites We identified the best model based on AIC and BIC criteria We used this model to predict cost curves with respect to scale, for different implementation scenarios
Determinants of unit cost variation OLS GLM Scale (log) -0.137*** -0.150*** (0.031) (0.035) Scale (log2) -0.007 -0.008 (0.014) (0.015) Rural=1, Urban=0 0.011 0.004 (0.079) (0.090) Private=1 , Public=0 0.275** 0.168 (0.125) (0.135) Hospital=1, Clinic=0 0.617*** 0.629*** (0.111) (0.119) Outreach=1 , Fix=0 0.278** 0.296** (0.121) (0.136) Year of data collection (2016 reference year) 0.146*** 0.186*** (0.029) Hospital*Private -0.529*** -0.535*** Constant 4.131*** 4.495*** Observations 220 Standard errors in parentheses *** p<0.01, ** p<0.05, *p<0.1
Determinants of unit cost variation (cont.) OLS GLM GDP 2016 (log) 0.562*** 0.492*** (0.062) (0.067) Constant 4.131*** 4.495*** Observations 220 Standard errors in parentheses *** p<0.01, ** p<0.05, *p<0.1
Scenarios (example) Country Urbanicity Ownership Type Facility Type Observed Unit cost per VMMC Kenya Urban Public Clinic 53.6 Hospital 37.5 Private 34.4 45.1 Rural 60.4 28.5 39.8 43.3 Namibia 87.8 South Africa 75.7 170.6
Produce scenarios
Cost curves per country
Summary We identified studies on VMMC unit costs and gathered data from 7 of them We are in the process of obtaining and standardizing data on ART, HTC, PMTCT, services for KP – about 1,600 observations so far We identified facility-level characteristics associated with unit cost variations Based on coefficients obtained from multivariate regression models we modeled cost curves for implementation scenarios
Thank you sbautista@insp.mx
Validate scenarios: South Africa Fix effects model GLM Random effects model
Validate scenarios: Kenya Fix effects model GLM Random effects model
Quantile regressions: Assess whether determinants influence unit costs differently in different quantiles of unit costs Mean p10 p25 p50 p75 p90 Rural -0.017 0.083 -0.011 -0.006 0.009 0.026 Private 0.291** 0.318* 0.255* 0.238* 0.235 0.167 Hospital 0.454*** 0.444*** 0.299** 0.275** 0.398*** 0.453* Outreach 0.040 0.091 0.034 0.072 -0.089 0.163 Scale (log) -0.238*** -0.228*** -0.216*** -0.233*** -0.227*** -0.186** Scale (log) 2 -0.026* -0.049** -0.035* -0.019 -0.021 -0.034 Year of collection (2016 ref.) 0.101** -0.095 0.019 0.104** 0.162*** 0.199** Hospital*Private -0.291* -0.316 -0.160 -0.087 -0.237 -0.346 Constant 4.013*** 2.388*** 3.319*** 3.998*** 4.646*** 5.004*** Observations 220
Cost curves per country 74% 100% 85% Note: % represents the percentage of observerved points contained by the scenarios 94% 70% 27% 55% 88%