An ecological study was performed to describe the use of mass drug administration for control and elimination of P. vivax malaria in Jiangsu Province, China (Figure  1A), and to explore the impact on malaria incidence. While MDA had been used since the 1950s, two periods identified by the provincial malaria programme as having detailed records available were focused on: a) 1973 to 1983 with MDA to target counties in the province, and b) 2000 to 2009 with focal MDA to villages and close contacts of index cases in select counties during a P. vivax epidemic. As individual level data on malaria cases and compliance with MDA and other interventions were not available, population level figures were utilized.

The goal of seasonal MDA with pyrimethamine and primaquine was to treat potential reservoirs of P. vivax blood and liver stage infection before the start of the malaria transmission season. Plasmodium falciparum had nearly been eliminated from Jiangsu Province by this time, accounting for fewer than 140 annual cases reported (Table  1).

Villagers were notified prior to the annual campaign. Community health workers and local public health officers administered medications daily by directly observed therapy (Figure  2A). MDA was administered before or at the start of the malaria transmission season that typically begins in April. From 1973 to 1976, seasonal MDA using pyrimethamine and primaquine was directed at all rural counties, and from 1977 to 1983, a stratified approach using chloroquine and primaquine was implemented (Table  2).

From 2000–2009, focal “spring treatment” was administered in counties on the western border as response to a P. vivax epidemic due to larger outbreaks in nearby Anhui Province. Initially, entire natural villages (population ~20 persons, the division below an administrative village) were targeted. As incidence declined, more focal approaches were taken based on the provincial mandate (Table  2). Exclusion criteria included: age < one year, pregnancy, serious heart, liver or kidney disease, fever, and history of cyanosis, systemic bleeding, or dark coloured urine.

MDA was conducted by local hospital and health bureau workers in the homes of targeted individuals. Informed consent was conducted verbally and through a handout. Villagers were instructed to have food before taking the medication. The administration of drugs was by directly observed therapy (Figure  2B) and documented by collecting patient signatures. Medication was stopped if serious adverse effects such as dark urine, cyanosis, or haemolysis occurred. The goal was for 85% of individuals infected in the previous year to complete the full course and for 80% of other individuals (family, neighbours, etc.) to complete the full course. All medication administration and data collection were to be completed by the end of March of that year [29].

Aggregate data were gathered from two counties (the administrative level below prefecture) in the western part of Jiangsu where focal MDA was implemented and detailed records were available: Sihong in Suqian prefecture and Xuyi in Huaian prefecture (Suquian and Huaian previously combined and known as Huaiyin prefecture) [31,32]. The data included monthly and annual incidence, as well as number of subjects that were targeted and successfully completed a MDA course. Data on adverse events and coverage of vector control measures were only available from Sihong. Annual population was available for all townships (administrative level below county) within Sihong and 22 of 28 in Xuyi [33,34]. For missing data, populations were imputed based on average township population that year. Published literature as well as county and provincial records were reviewed for any data on haemolysis or other adverse effects relating to focal MDA.

For both time periods, gross domestic product (GDP) data from China were collected from the World Bank and deflated to year 2000 US$ to account for inflation. All monthly rainfall data were collected from the International Research Institute for Climate and Society [35]. For the 1978 to 1983 period, monthly rainfall data were calculated as the average from 14 weather stations throughout Jiangsu Province. For the 2000 to 2009 period, average monthly rainfall data were collected from two weather stations closest by coordinates to the selected counties.

Data were extracted from paper records with support of a translator, and then double entered into Microsoft Excel (Microsoft Corp, Seattle, USA). Data were cleaned and analysed in R (Foundation for Statistical Computing, Vienna, Austria) and Stata 11.0 (Stata Corp, College Station, TX, USA).

Total MDA coverage was defined as per cent of the total population covered by MDA (total population coverage). This value did not take into account the fact that only high-risk or target villages and individuals received MDA. Therefore, target population coverage, or per cent of the target population covered by MDA for the 2000-2009 dataset (data not available for 1973-1983 dataset) was also presented. Also, as a functional measure that has relevance to operations and implementation, particularly for focal MDA where only villages or household members and neighbours are treated, the ratio of population receiving MDA to incident case was also calculated.

Given that incidence rate appeared to vary over the study time period, trends were analysed using joinpoint regression analysis, a statistical method that fits a series of joined straight lines to rates over time and chooses best-fitting points or joinpoints where the rate of increase or decrease changes significantly [36]. The grid search method was used to detect segments best describing the data. This method creates a “grid” of all possible locations for joinpoints and tests the error sum of squares at each one to find the best possible fit. Joinpoints were identified as years in which a significant change in trend was detected over the study period. Annual per cent change with 95% CI for each segment are reported to describe linear trends by period.

For the 1973-1983 Jiangsu Province dataset, the association between annual total population MDA coverage and incidence was estimated using population average negative binomial models. To control for background incidence trend as a potential confounder, an interaction term for time period was included based upon segments identified in the joinpoint analysis: early (1973-1976) vs later (1977-1983). A Poisson model was considered but not used as the negative binomial distribution provided a better fit for the data given the high amount of over-dispersion. The negative binomial analysis modelled monthly counts (monthly cases of P. vivax) as the outcome, taking into account prefecture population as an offset. Regression coefficients were computed as incidence rate ratios (IRRs). Total annual population MDA coverage was the primary exposure variable and other covariates controlled for including: annual chemoprophylaxis coverage, net coverage (defined as the number of nets distributed per population), rainfall lagged by one month, and deflated GDP. Drug salts were not included in the model due to non-significant effect and collinearity with chemoprophylaxis and bed nets. Generalized estimating equations with an autoregressive correlation structure were used to control for serial correlation within prefecture time series. Monthly and prefecture dummy terms were used to control for seasonality effects beyond rainfall and unobserved prefecture fixed effects, respectively. To control for potential spatial endogeneity between prefectures, the model included anomalies for intervention terms (computed as the difference between annual coverage and the local mean coverage over the study period for each prefecture). Rainfall and GDP values were standardized to represent one standard deviation (by subtracting the mean and dividing by the standard deviation).To evaluate the association between MDA and incidence during different time periods, a binary term for early (1973-1976) vs. later (1977-1983) years was interacted with the MDA term. Confidence intervals (CI) of 95% are presented based upon robust standard errors. Due to low total population coverage and incidence, the model was not applied to the 2000-2009 county datasets.

Human subjects’ approval was not indicated because the surveillance data utilised was existing, aggregate and without personal identifiers. The analysis protocol was reviewed by the Centers for Disease Control and Prevention IRB and received non-research determination.

In almost all prefectures, intermittent chemoprophylaxis was used during the transmission season from June to October. The regimen consisted of either pyrimethamine 50 mg every two weeks or chloroquine 300 mg every ten days. Per cent population coverage of chemoprophylaxis increased from 3.3% in 1973 to 9.8% in 1976, then declined to 1.1% by 1983. At the height of the programme, up to 44 million cumulative doses were administered a year. From 1976 to 1983, a few prefectures also distributed salt fortified with pyrimethamine. From June to September, each person received a monthly average of 400 g salt containing 157.5 mg pyrimethamine [37]. Population coverage was highest at 2.4% in 1982. Bed nets were utilized beginning in 1978, however population level coverage was very low. Indoor residual spraying was not part of the provincial malaria strategy.

Treatment for clinical malaria cases varied by county. The most commonly used regimen was chloroquine on days 1-3 (1.2 g total) with primaquine on days 1-4 (22.5 mg daily). One month later patients were treated with primaquine for another four days (22.5 mg daily) for a total primaquine dose of 180 mg. Another regimen used was chloroquine on days 1-3 (1.2 g total) with primaquine on days 1-8 (22.5 mg daily, 180 mg total). From 1973 to 1977, a five-day regimen was also used in some counties, chloroquine on days 1-3 (1.2 g total) with primaquine on days 1-5 (22.5 mg daily, 112.5 mg total). Subjects who were infected in the previous year were also individually targeted for seasonal MDA.

In the setting of large scale MDA, the incidence of P. vivax in Jiangsu Province declined from 113.6 in 1973 to 2.1 per 1,000 population in 1983 (Table  1, Figures  1B- 1D, 3). Joinpoint analysis on the aggregate provincial dataset identified two distinct periods: 1973-1976 when annual per cent change was -56.7% (95% CI -75.5 to -23.7%) and 1976-1983 when annual per cent change was -12.4% (95% CI -24.7 to 2.0%). At the prefecture level, no more than two distinct periods were identified, with joinpoint breaks occurring from 1976 to 1978. In most prefectures, annual per cent changes in incidence were negative (Table  3).

The population average negative binomial model confirmed that time period significantly impacted the relationship between total population MDA coverage and incidence: monthly incidence was significantly lower in the later period compared with the earlier period (IRR = 0.19; 95% CI 0.11 to 0.33), after controlling for potential confounding factors including chemoprophylaxis, bed nets, rainfall, and GDP (Table  4). Between 1973 and 1976, higher total population MDA coverage was associated with significantly lower monthly incidence (IRR = 0.98; 95% CI 0.97 to 1.00) after controlling for confounders, but in the later period, it was associated with higher incidence (IRR = 1.05; 95% CI 1.03 to 1.06). Other potential confounders were not significantly associated with monthly incidence.

Adverse events from MDA were not systematically monitored as part of the programme, but five reports documenting G6PD deficiency-related severe adverse events from primaquine were identified (Table  5). Two case series were recorded in the setting of large-scale campaigns, one from 1975 and one from 1976. There were two cross-sectional studies, one among family members of patients with severe G6PD deficiency and one among high-risk groups. Additionally, there was a case report from 1979.

In the two case series, drugs were administered to 254,910 residents in Jinhu county and 44,589 residents of Liuhe county. The incidence of acute haemolysis was 3.5 and 9.3 per 100,000 population, respectively. Symptoms were typical for acute haemolytic reaction. Among the 40 patients from Liuhe county, G6PD deficiency testing was performed using the methaemoglobin reductive test, which had a sensitivity of 65% for detection of patients with acute haemolysis. In both studies, there were no deaths and all patients recovered with drug discontinuation and supportive care.

In one of the cross-sectional studies, methaemoglobin testing was performed in 94 family members of three patients who suffered from haemolysis after primaquine and 59.6% tested positive. In another study, methaemoglobin testing was performed in groups hypothesized to have higher prevalence of G6PD deficiency. The prevalence of a deficiency was 5.7% in a village with a high incidence of acute haemolysis versus 0.7-1.0% in the other groups (a village with past P. falciparum cases and a village with members of the Hui minority group) and 0% in a control group of patients with liver disease. The higher prevalence in the village with a high incidence of acute haemolysis was thought to be due to a genetic predisposition in this community.

Unlike in the 1973-1983 time period, other prophylactic drugs were not utilized. In Sihong county, there was some use of vector control measures. Insecticide-treated bed nets covered 0 to 3% of the population between 2000 and 2004, 10% in 2005, and 4 to 8% between 2006 and 2009. Indoor residual spraying was performed in 2007 and 2009 in select townships. Coverage was recorded as 109.5 kg sprayed over 47,706 sq m in 2007 and 231 kg sprayed over 109,125 sq m in 2009. No information was available on proportion of structures sprayed. There was no recorded use of vector control measures in Xuyi. During this time period, clinical cases of P. vivax presenting to health facilities were treated with chloroquine for three days (1.2 mg total dose) and primaquine for eight days (180 mg total dose).

Although the peak incidences were low at baseline, 0.79 and 0.29 per 1,000 population in Sihong and Xuyi, respectively, incidence declined further in the setting of focal MDA (Figure  4). Sihong and Xuyi counties saw declines the year after focal MDA was started. Low incidences were maintained even as focal MDA coverage declined. Joinpoint analysis identified two distinct periods in Xuyi and one in Sihong (Table  3). In Xuyi, the break correlated with a change in total population coverage of focal MDA (Figure  4). Annual per cent changes were negative and statistically significant in all segments. Temporally, other factors did not appear to contribute to decreasing incidence. Gross domestic product increased over the study period but the rate was steady. Rainfall was highest in 2000, 2005 and 2007 (Additional file 1: Figures S1b and S2b). Vector control measures were used in Sihong, but the periods of highest coverage do not correspond with declines in incidence.

In Sihong, the number of subjects with adverse reactions was recorded in 2003 and 2007. Adverse reactions were few, occurring in five subjects (2.1/1,000) in 2003 and two (4.3/1,000) in 2007. Symptoms included haemolysis with primaquine, dizziness, gastrointestinal upset, nausea, and vomiting with chloroquine. Safety data were not available from Xuyi.