Health facility delivery service utilization
A statistically significant increase in the number of deliveries in the health facilities was identified; this number increased from 234,601 before policy implementation to 303,705 after policy implementation, representing a 29.5% increase (p < 0.05; Table 1).
Table 1 Total Deliveries in the different levels of health facilities The results of the analysis of quarterly deliveries in the 77 health facilities indicated a decreasing trend in deliveries (slope = − 13.131, p = 0.00) during the 24 months preceding implementation of the policy. Thus during the 24 months before the intervention, no significant change was identified in the number of facility-based deliveries. A significant increase, however, in the quarterly number of facility-based deliveries (slope = 124.90, p < 0.01) in the 77 health facilities was identified after policy implementation (Table 2).
Table 2 Quarterly Patterns in Health Facility Delivery Service Utilization A closer look at the delivery service utilization trends identified during the 6 months before policy implementation indicates the presence of a decreasing trend in the utilization of facility-based delivery services (slope = − 124.90, p = 0.02). During the 6 months after policy implementation, this trend reversed, and a significant increase in the number of deliveries in health facilities was observed (slope = 111.77, p < 0.001).
Both the stationary and the traditional R2 tests yielded a value of 0.73, implying that 73% of the model was explained by the policy intervention. In addition, a root mean square error (RSME) value of 384.22 was identified, suggesting that a large portion of the variability observed in the number of deliveries could be explained by the predictive model. The mean absolute percentage error (MAPE) value of 7.25 indicated that the values predicted using the policy implementation model were, on average, within 7.25% of the actual values (Additional file 1).
Both the stationary R-squared and traditional R-squared values for all 77 health facilities were 81.5% (P = 0.15; Additional file 2). The stationary R-squared and traditional R-squared values varied from 73.1% to 43.7% across various categories of health facilities. This finding indicated that although policy implementation resulted in a remarkably higher number of facility-based deliveries, this intervention had a non-uniform effect on delivery service utilization across the 77 health facilities (p = 0.15).
Maternal mortality ratio
A nonsignificant decrease in the ratio of maternal mortality in the 77 health facilities was identified, with the mortality ratio decreasing from 258.3/100,000 live births to 237.1/100,000 live births (p = 0.07) following policy implementation (Table 3). It is only in the rural areas that a significant decline in maternal mortality ratio was recorded.
Table 3 Maternal Mortality Ratios The ARIMA model parameters for the pre-intervention slope that was calculated using data from the 77 health facilities showed a nonsignificant decrease in the rate of quarterly maternal mortality (slope = − 1.64, p = 0. 20) during the 24 months preceding user fee removal. During the 24 months after free maternity health care services were first offered, a significant increase in the rate of quarterly maternal mortality was observed in the health facilities under consideration (slope = 3.49, p = 0.01). This finding indicated that the free maternal health care policy did not have a significant effect on facility-based maternal mortality ratios (Table 4).
Table 4 Quarterly Trends in Maternal Mortality Ratio Both the stationery R-squared and traditional R-squared (R2) values for the model were 0.126, implying that only 12.6% of the variance observed in maternal mortality ratio could be explained by the free maternal health care policy intervention. The RSME value of 112.67 indicated that the interrupted time series model was reliable in predicting maternal mortality trends. The calculated MAPE indicated a 33.77% variation from the model prediction following the policy intervention (Additional file 3).
Overall, the Ljung-Box test statistics were not significant for the 77 health facilities (p = 0.54). Significant Ljung-Box test statistics were identified for level 5 health facilities only (30.64, p < 0.05). The stationary R-squared and traditional R-squared values indicated that only a minimal decline in maternal mortality ratio occurred in the 77 health facilities (0.19), with the greatest decline in maternal mortality ratio identified in the level 6 health facility (0.56; Additional file 4). The maternal mortality ratios demonstrated a decreasing trend in the health facilities but exhibited marked although not consistently uniform seasonality. Thus, the free maternal health care policy intervention had a random and nonsignificant effect on maternal mortality ratios across all health facilities.
Neonatal mortality rate
A nonsignificant decline in neonatal mortality rates was identified, with rates decreasing from 23.3/1000 live births to 22.9/1000 live births (p = 0.14) following policy implementation (Table 5).
Table 5 Neonatal Mortality Rates The pre-intervention slope indicated a nonsignificant decreasing trend in the quarterly neonatal mortality rates across all health facilities over the course of the 24 months preceding policy implementation (slope = − 0.09, p = 0.24). Implementation of the policy did not significantly affect neonatal mortality rates during the first 24 months following policy implementation (slope = 0.12, p = 0.10; Table 6).
Table 6 Quarterly Trends in Neonatal Mortality Rate Only 32.90% of the minimal and nonsignificant change observed in the neonatal mortality rates could be attributed to policy implementation. The RSME and MAPE values for this model were 688.10 and 17.96, respectively (Additional file 5).
The general fit of the model for all 77 health facilities was not significant (p = 0.06). Both the stationary R-squared and traditional R-squared values indicated that only 10.5% of the variation could be explained by the model, implying that policy implementation was associated with only a minimal difference in neonatal mortality rates when compared with baseline figures (Additional file 6).