Country context

Routine distribution of ITNs targeting children under 5 years old and pregnant women started in 2003 in 20 pilot districts and was scaled up nationwide during 2004–2010. During 2006–2009, >4.5 million ITNs were distributed through antenatal care (ANC), Child Health weeks, voucher schemes, and commercial sources. However, the strategy was replaced with a long-lasting insecticidal net (LLIN) mass campaign, starting at the end of 2010 with the goal of achieving universal coverage in all ages at a ratio of one net per two persons. During 2010–2012, >12 million LLINs were delivered, and another 12.3 million LLINs during 2013–2015. In 2012, the AngloGold Ashanti Malaria Control Programme (AGAMal) implemented indoor residual spraying (IRS) in seven districts, expanding to 22 districts in five regions (Upper West, Upper East, Ashanti, Western, Central) in 2015. PMI supported five districts for IRS in the Northern region. In total, in 2015, 4.7% of the entire population in the country benefited from IRS. This population also received LLINs.

Parasitological testing of patients of age 5 years and above was primarily limited to hospitals. Since 2010, confirmation of all ages with either microscopy or rapid diagnostic tests (RDTs) has been introduced. Health centres and health posts use RDTs as primary diagnostic tool while higher facilities use microscopy, reserving RDTs for emergency cases, and for situations where microscopy is not functional or caseloads exceed microscopy diagnostic capacity. Therefore, only records of microscopic testing from the hospitals were used for this study.

The country changed its first-line anti-malarial treatment from chloroquine to artesunate–amodiaquine (AS–AQ) in 2004 and added artemether–lumefantrine (AL) and dihydro-artemisinin–piperaquine (DHAP) as alternatives in 2009. Quinine was the principal medicine for severe malaria until 2010, when injectable artesunate was adopted as the drug of choice. The Affordable Medicines facility for Malaria (AMFm), launched in 2011, increased access to ACT in the private sector from 31% in 2010 to 83% in 2011 [3].

For community-based intervention, between 2005 and 2012, nearly 2218 CHPS have been established in the country. The health staff in the CHIPS diagnose with RDTs and treat malaria and assist delivery at the community. However, presumptive treatment of malaria may occur for logistic reasons, when RDTs stock-out.

Prior to 2007, patients of all ages were charged for malaria diagnosis and treatment, which was a major barrier to accessing basic health services for majority of the population. To address the problem of access, the NHIS was initiated in 2003 and became operational in 2005 to subsidize maternal and child health services, which became free in 2007. In the NHIS, urban and rural populations pay a premium of about US$17 and US$11 per year, respectively, and when the cost is beyond what NHIS cover, patients cover the difference. For people whose inability to pay is certified by local authorities, the Government covers all. Despite these provisions, the coverage of the NHIS remains at 38% owing to financial challenges to sustain the scheme, identification of the poor and vulnerable, identification (ID) card management, quality of care, and slow information system [4].

Data for intervention coverage

Long-lasting insecticidal nets and IRS data were obtained from district records. The proportion of the population potentially protected by LLINs was calculated for each year, assuming each LLIN covered 1.8 persons and lasted 3 years. Prospective district population was derived from the 2010 Ghanaian census, using the United Nations growth rate for Ghana [5, 6]. To measure ACT availability, number of months-ACT stock-out was obtained from records of hospital medical stores. Stock-out was defined when a hospital had no ACT in stock for more than 7 days in a month.

Data for malaria morbidity and mortality

A WHO standardized protocol (unpublished) was used for hospital data collection. A total of 88 hospitals in three epidemiological zones were randomly sampled (28 in savannah, 30 in forest, 30 in coastal). Although the design was to select 30 representative hospitals from each zone, the savannah had only 28 hospitals and all were sampled. Data abstraction teams consisting of two persons visited each hospital for 2 days.

Monthly summary reports were used as the main source of records for: (i) outpatient all-cause consultations and malaria cases; (ii) inpatient all-cause and malaria admissions, all-cause deaths, malaria deaths, anaemia admissions, and anaemia deaths; and, (iii) laboratory records for microscopically tested and positive cases. Where monthly summary records were missing, register books were used as primary source. Data were collected for two age groups: all ages and under 5 years of age.

Analysis of microscopic tests and confirmed cases was possible only for all ages as records of microscopic results could not be disaggregated by age, outpatient or inpatients. Microscopically confirmed malaria cases, malaria admissions and malaria deaths were the main interest of the analysis. A confirmed malaria case for this study was defined as a blood slide positive for malaria. Malaria admission was defined as a microscopically confirmed inpatient case, diagnosed at discharge as severe malaria. However, some malaria admissions were admitted based on clinical signs suggestive of severe malaria. A malaria death was defined as death attributed to malaria among the admitted cases. All anaemia admissions were assumed to be severe anaemia (<7 mg/dl). The test positivity rate (TPR) was computed by dividing number of positive slides by the total number of slides examined.

To control for other factors that might influence the observed trends, the following additional indicators were computed: (i) non-malaria outpatient consultations (all-cause outpatient cases minus outpatient malaria cases); (ii) non-malaria admissions (all-cause admissions minus malaria admissions); and, (iii) non-malaria deaths (all-cause deaths minus malaria deaths). In addition, data on admitted patients with or without health insurance were extracted from the DHMIS-2 to assess the difference in disease trends among those insured and non-insured. These allowed the assessment of whether changes in malaria cases and deaths were truly attributable to interventions or to changes in healthcare-seeking behaviours or other factors that affect all diseases in a similar way. In addition, the trends of child mortality from population-based surveys for Ghana were also reviewed and triangulated with the trends observed in the hospitals.

To control for the potential effect of climatic changes in trends of malaria during pre- or post-scale-up periods, data on monthly rainfall in mm, minimum and maximum temperature in  °C by region, were obtained from the Ghanaian Meteorology Department.

Statistical analysis of time trends

For filling missing monthly data, either of the following approaches were applied: (i) if the missing data were in the 1st year, then average the consecutive 2 years of the same month, (ii) if the missing data were in the last year, then average the two preceding years of the same month; and, (iii) if the missing data were in between 2 years, running averages of the preceding and consecutive year for the same month. Because the data points end in June 2015, monthly data were aggregated biannually (two periods for each year) making a total of 21 periods covering January 2005–June 2015.

Assuming all other interventions constant, the effect of an LLIN mass campaign since 2011 was evaluated by comparing the disease trends of the pre-scale up years (2005–2010), i.e., period 1–11 with that of period 12–21 (post-scale up years 2011–2015). Changes on trends were estimated as relative per cent change by comparing the observed value and predicted value at period 21 assuming a continuation of the pre-scale up time trend throughout period 21 if there were no interventions (counterfactual trend). The interpretation of the predicted relative change in percentage at period 21 therefore takes into account the trends (slop) of all the values during 2011 and 2015. This was done using a segmented regression model of an interrupted time series with breakpoint of period 11 [7], an approach that has increasingly become useful to evaluate impact of public health interventions [8, 9] including malaria [10, 11]. The 95% confidence intervals (CI) around effect estimates were computed using the CI around the regression coefficient. A percentage difference with a CI that does not include zero in the range was considered a significant change (P < 0.05). This model corrects for autocorrelation [12, 13] and adjusts the change estimate allowing for: (1) possible time trend of the indicator during the pre-scale-up period, (2) a possible immediate drop or rise of the indicator following the start of the campaign; and, (3) an effect of the campaign on the post-scale up at a given period. In addition, a stratified analysis using segmented regression was applied to compare changes in disease trends in IRS districts (Northern region) and non-IRS districts.

Interventions

Mapping of the interventions is given in Fig. 1. The proportion of population of all ages potentially protected by LLINs increased from nearly 25 in 2005 to 86% in 2015 (Fig. 2). The type of LLIN used in mass and routine distribution during 2005–2015 was a mixture of Permanet^® Olyset^®, Interceptor, and Netprotect. The percentage of the hospitals with available stock of all forms of ACT was 100% during 2006–2012, and >95% during 2013-2015 (Fig. 2). Availability of quinine in stock was 100% during 2005–2012 while it was between 80 and 90% during 2013–2015. After its introduction in 2010, availability of artesunate injectable increased to 83% in the first half of 2015.

Trends on the malaria outpatient cases, inpatient and deaths

Among all ages, the number of suspected malaria cases declined from a 6-months average count of 635,161 during 2005–2010 (period 1–10) to 581,906 in the first half of 2015 (period 21), with an observed decrease of only 8%, but significant predicted decrease of 57% (95% CI, 47–66%) during 2011–2015 compared to the trend of 2005–2015 (assuming continuation of the trend at the pre-scale-up level). The number of microscopically confirmed cases had little change in observed values but a significant predicted decrease of 53% (28–69%), while the number of microscopic tests had an observed increase and little change in predicted values. The TPR in all ages declined by 30% (observed) and a predicted decrease of 41% (52–75%) nationwide during the same period. Malaria admissions showed 52% increase in observed trends and insignificant predicted change, 39% (0–63%). Malaria deaths fell by 49% in observed and predicted trends significantly by 65% (52–75%) (Table 1; Fig. 3a–d). The number of anaemia inpatient cases showed an observed increase of 126% but insignificant predicted trend. Anaemia-related deaths increased by 47% with insignificant predicted changes. The proportion of outpatient malaria cases, malaria admissions and malaria deaths of all-cause conditions in the hospitals showed an observed decrease of 36,17 and 52% respectively and a significant predicted decline of 32% (21–42%), 41% (23–54%) and 61% (48–70%), respectively (Table 1; Fig. 3e).

No of HFs	Indicator	All ages				Under 5
		Baseline	Observed values	Observed relative change (%)	Predicted relative change compared to baseline (%)	Baseline	Observed values	Observed relative change (%)	Predicted relative change compared to baseline (%)
		Ave (2005−2010) P1	2015 P1	2015 P1	2015 P1	Ave (2005−2010) P1	2015 P1	2015 P1	2015 1
88	Outpatient malaria cases	635,161	581,906	−8	−57 (−66 to −47)†	145,419	155,677	7	−50 (−63 to −32)†
	Malaria admissions	60,070	91,425	52	−39 (−63 to 0)†	26,274	38,857	48	−46 (−75 to 19)
	Malaria deaths	1229	632	−49	−65 (−75 to −52)†	569	234	−59	−70 (−82 to −49)†
	Anaemia inpatients	10,611	24,004	126	1 (−68 to 221)	5082	9245	82	−17 (−75 to 179)
	Anaemia deaths	456	672	47	38 (−19 to 135)	178	158	−11	−24 (−63 to 59)
70	Tested (Mic)	247,750	374,394	51	−22 (−54 to 34)	85,514	129,092	51
	Positive (Mic)	87,610	92,383	5	−53 (−69 to −28)†	29,859	32,476	9
	Test positivity rate (mic)	35	25	−30	−41 (−57 to −19)†	36	25	−29
88	Proportion of outpatients	36	23	−36	−32 (−42 to −21)†	54	35	−35	−30 (−37 to −21)†
	Proportion of inpatients	25	21	−17	−41 (−54 to −23)†	53	32	−39	−49 (−57 to −40)†
	Proportion of deaths	15	7	−52	−61 (−70 to −48)†	28	12	−56	−59 (−72 to −40)†
	Non-malaria outpatient	1706,037	2439,358	43	−8 (−28 to 17)	237,858	409,470	72	−28 (−48 to 1)
	Non-malaria inpatients	174,948	344,964	97	24 (−11 to 72)	23,071	80,890	251	94 (−7 to 303)
	Non-malaria deaths	7175	8340	16	−1 (−23 to 28)	1497	1697	13	−8 (−36 to 32)

CI with symbol “†” is statistically significant

In children under 5 years old, total malaria outpatient cases (both confirmed and presumed cases) had little change in observed values but a significant predicted decrease of 50% (32–63%). Malaria deaths had an observed 59% decrease and a significant predicted decline of 70% (49–82%). The number malaria admissions had an observed increase of 48% and insignificant predicted change of 46% (19–75%); the proportion of outpatient malaria cases, admissions and deaths of all-cause conditions showed an observed decline of 35, 39 and 56% respectively; and a significant predicted decline of 30% (21–37), 49% (40–57%) and 59% (40–72%) (Table 1). The number of anaemia inpatient cases showed an observed increase of 82% but insignificant predicted trend. Anaemia-related deaths decreased by 11% with insignificant predicted changes.

Concomitantly, the non-malaria outpatient cases, non-malaria inpatient and non-malaria deaths in both all ages and children under five generally increased or remained unchanged during the same period (Fig. 3; Table 1).

Population based mortality

All-cause mortality for children under 5 years of age in household surveys declined from 105 per 1000 live births during 2004–2008 [16], to 82 per 1000 live births during 2009–2011 [15], and to 60 per 1000 during 2010–2015 [16] representing at least a decline of 43% between 2008 and 2014. Five-year cumulative all-cause deaths in under 5 years in the 88 hospitals declined from 7939 during 2005–2010 to 5340 during 2011–2015 (33% decline). During the same period, hospital malaria deaths declined by 49% (1173 to 594). Total all-cause child deaths at population level decreased from 348,440 to 258,673 during the same period (26% decline). Deaths in the 88 hospitals represented only 1% of all the deaths at population level attributed from the household surveys (Fig. 4).

Trends among the insured and non-insured by NHIS

Nationwide hospital data from the DHIMS-2 were used to discern the trends of all-cause admissions among the insured and non-insured. All-cause admissions among the insured in all ages increased by 99% while it decreased by 21% among the non-insured patients (Fig. 5).

Spatial differences

Difference in IRS and non-IRS districts

Effect of rainfall anomalies and seasonality

There were no noticeable anomalies in rainfall and temperature that could plausibly affect the normal seasonal trends of malaria cases and deaths (Fig. 6a). Mean minimum and maximum rainfall was 82 and 118 mm, respectively. The mean minimum temperature during pre- and post-scale-up was similar, 22.0 °C (21.5–22.8 °C) and 22.9 °C (22.2–23.8 °C), respectively; the mean maximum temperature was 31.8 °C (30.3–33.3 °C) and 31.6 °C (27.8–33.8 °C), respectively (Fig. 6b). A time-series regression on the effect of the three climatic factors on malaria deaths and TPR, as an example, showed little effect from variations in rainfall and maximum temperature. Minimum temperature appeared to have significant effect on reducing the trends of both indicators (malaria deaths and TPR) controlling for effect of time (P < 0.05) (Table 4).

	Coefficient.	Std. err.	t	P > t	(95% CI)
					Low	High
Malaria deaths
Rainfall	0.000181	0.002017	0.09	0.930	−0.00414	0.004507
Min temperature	−0.17246	0.041664	−4.14	0.001	−0.26182	−0.0831
Max temperature	0.003815	0.022157	0.17	0.866	−0.04371	0.051337
Constant	10.90383	0.604018	18.05	0	9.608344	12.19933
Test positivity rate
Rainfall	0.003097	0.003024	1.02	0.364	−0.0053	0.011492
Min temperature	−0.16048	0.044244	−3.63	0.022	−0.28332	−0.03764
Max temperature	−0.00271	0.019038	−0.14	0.894	−0.05556	0.050149
Constant	7.395645	0.677037	10.92	0	5.515888	9.275402

Limitations

Caution has to be made when interpreting the results of such health facility-based data often affected by changes in service utilization and reporting completeness [35]. The analysis in this study addressed this problem through data collection from selected 88 health facilities (representing three epidemiological zones) with complete reporting and parasitological testing covering the study periods. Other limitations include: (i) inability to assess the trends in peripheral facilities such as health centres and CHIPs as testing with RDTs in these facilities started only after 2010; (ii) inability to assess the trends in private health facilities; (iii) incomplete data on RDT testing in the hospitals. Although the use of RDTs in the hospitals is limited only in the absence of microscopist and confirmation of highly suspicious severe cases, such usage may have affected the testing rate or patterns of use of microscopic diagnosis; and, (iv) inability of the data to show disaggregated trends of insured and non-insured malaria admissions.