Mobility and malaria incidence in Indonesia’s low-endemic areas during the COVID-19 period Authors: Risalia Reni Arisanti, Ahmad Watsiq Maula, Riris Andono Ahmad, Adi Utarini Published: 2026 ABSTRACT BackgroundHuman mobility significantly impacts the transmission and persistence of malaria, even in low-endemic areas. The COVID-19 pandemic caused significant shifts in mobility patterns across Indonesia. This study examines the association between population mobility and malaria incidence in low-endemic Indonesian districts during the COVID-19 period.MethodsThe study was conducted from March 2020 to May 2022 in four districts, i.e., Kulon Progo and Purworejo (Java Island) and Tanah Laut and Tanah Bumbu (outside Java Island). Data were collected from district health office reports on COVID-19 cases, mobility data from Facebook, malaria testing, and case reports from the district health office’s electronic system. Multivariable Poisson regression was used to examine the association between mobility and malaria incidence, adjusting for district location, endemicity of mobility origin, malaria testing intensity, and COVID-19 burden.ResultsA total of 480,575 mobility events were examined, showing that mobility peaked in 2021 outside Java Island, with Tanah Bumbu exhibiting the highest mobility rate (36.6%). In Java, most mobility came from elimination areas (> 80%), while outside Java areas showed mobility from the low-endemic regions (48.5% and 87.9%). In Purworejo, mobility fluctuated in 2020, with peaks in March and May 2021, correlating with a rise in malaria cases. In Kulon Progo, mobility increased only slightly, but malaria cases spiked in late 2021 and early 2022. No clear pattern was found between mobility and malaria cases in Tanah Bumbu and Tanah Laut. Malaria examination rates dropped significantly during the early months of the pandemic. The incidence of malaria outside Java was 64% lower than in Java, and increased mobility was associated with higher malaria incidence. In the pooled multivariable Poisson regression, mobility, malaria screening, and COVID-19 positivity were associated with malaria incidence [IRR 1.25 (95% CI 1.21–1.30), IRR 2.33 (95% CI 2.22–2.45), and IRR 1.12 (95% CI 1.08–1.17), respectively]. However, descriptive analysis showed no clear temporal association between mobility trends and malaria cases outside Java, suggesting the relationship is context dependent.ConclusionsMobility remains a crucial factor in malaria control, although its relationship is complex and context-dependent, influenced by local patterns of endemicity. These findings are associated in nature and do not causal effects. Targeted interventions during such periods are essential to prevent a resurgence of the disease. Specifically, the National Malaria Control Program should consider integrating real-time digital mobility indicators into SISMAL to generate mobility-informed risk alerts for districts experiencing elevated inflows from endemic areas.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12936-026-05945-6. Background Human mobility has long been recognised as a primary factor in the transmission and persistence of malaria. Individual movement between high- and low-endemic areas enables malaria parasites to be reintroduced into previously controlled or eliminated regions, thereby increasing the likelihood of re-establishing local transmission [1, 2]. Imported infections increase the risk of outbreaks and delay progress toward elimination in low-endemic environments where malaria transmission is unstable, and community immunity is low [1]. Understanding the dynamics of movement is thus critical to establishing targeted malaria elimination efforts, particularly in countries such as Indonesia, where the goal is to eliminate malaria by 2030. Indonesia presents a unique context, with heterogeneous malaria endemicity across its archipelago [3]. By 2022, 72.3% of districts had been certified malaria-free, mainly in Java Island. The remaining districts are highly endemic (28), moderately endemic (27), and low-endemic (87) [4]. The geographical diversity of malaria risk means that frequent mobility between high- and low-risk areas, such as seasonal migration, labour-related movements, and travel to forested areas, can perpetuate malaria transmission [2]. Previous studies have shown that occupational groups, such as miners, forest workers, and mobile migrants, often serve as reservoirs, introducing malaria parasites into receptive but low-transmission regions [5, 6]. These dynamics underscore the vulnerability of low-endemic districts to case reintroduction through human movement. The COVID-19 pandemic created unprecedented disruptions in human mobility and health services. Nationwide restrictions, lockdowns, and changes in economic activity substantially altered patterns and reduced the frequency of movement across the country [7]. At the same time, malaria surveillance and diagnostic services were interrupted, particularly during the early stages of the pandemic and the Delta wave in mid-2021 [8]. These dual pressures, reduced access to health services and shifting mobility flows, may have linked malaria epidemiology in vulnerable, low-endemic districts [9, 10]. Some regions experienced reduced malaria transmission due to restricted movement, while others faced risks of delayed diagnosis, underreporting, and localised outbreaks [10]. Measuring human behaviour at scale and in near real time presents numerous challenges and constraints. Historically, behavioural patterns were mainly derived from self-reported survey data, which is resource-intensive and subject to reporting bias. More recent systems rely on low-resolution mobile phone records; however, privacy concerns and limited accessibility constrain their use [11]. Online platforms such as Twitter, YouTube, Facebook, and Instagram generate massive, diversified, and cost-effective data streams, opening new avenues for behavioural observation [12]. Location-sharing platforms, such as geotagged Facebook data, provide academics with unparalleled access to direct records of human mobility [13]. During the COVID-19 epidemic, when resources are limited, the use of geotagged social media metadata has become increasingly important. Facebook provides free user data samples that are widely used in Indonesia, with district-level mobility information primarily from young and older people. Such findings are significant for infectious diseases such as COVID-19 and malaria, in which asymptomatic individuals contribute to transmission. Furthermore, infections are more common among adults in low-endemic malaria areas, making these data particularly useful for estimating migration in such settings. Despite this evidence, the specific association between pandemic-related mobility changes and malaria incidence in low-endemic elimination-phase settings, using near-real-time digital mobility data, remains poorly characterised for Indonesia. This study addresses this evidence gap by examining associations between Facebook-derived mobility indicators and malaria incidence across four low-endemic Indonesian districts during the COVID-19 period. Understanding the association between mobility and malaria cases is critical for informing malaria elimination strategies, ensuring surveillance resilience during health emergencies, and preventing resurgence in low-endemic regions. This study examines the association between COVID-19–related mobility changes and malaria incidence in low-endemic districts of Indonesia, which remain vulnerable to the reintroduction of cases and outbreaks. Using mobility data alongside malaria case and testing records, we provide evidence on how changes in human mobility associate with malaria trends across Java and outside Java Island. Methods Study site Indonesia is characterised by low-endemic malaria transmission across most of the country, particularly in Java, Bali, Sumatra, and large parts of Kalimantan and Sulawesi, where malaria incidence is low and often limited to sporadic or imported cases. Java alone is home to more than half of Indonesia’s total population, making the country predominantly low endemic in terms of population exposure. In contrast, moderate to high transmission persists in eastern Indonesia, especially in Papua and West Papua, which continue to account for the most national malaria cases [14]. We selected four districts based on the following criteria: (1) low-endemic malaria classification during the study period; (2) representation of two geographic regions (Java and outside Java); (3) diversity of ecological socioeconomic contexts (semi-urban transit corridor, coastal agricultural, and mining-extraction settings); (4) availability of complete SISMAL malaria surveillance data for the 2019–2022 period. Figure S1 (Supplementary) presents a map of Indonesia showing the geographic location of all four districts and the national malaria endemicity classification (elimination, low-endemic, moderate-endemic and high-endemic) at the district level. Two study sites were located in Java: Kulon Progo (Yogyakarta Province) and Purworejo (Central Java Province), while two were outside Java: Tanah Laut and Tanah Bumbu (South Kalimantan Province). In Java, both selected districts are located in densely populated areas. Kulon Progo, which borders the southern shore of Yogyakarta, is a semi-urban district with mixed agricultural and peri-urban populations. Purworejo is located near Yogyakarta and serves as an essential transit corridor, connecting the provinces of Central Java and Yogyakarta. The other two sites were located outside Java, in South Kalimantan Province, specifically in the Tanah Laut District and Tanah Bumbu District. These districts represent areas with distinct ecological and socio-economic conditions compared to Java. Tanah Laut is a coastal district with substantial agricultural and plantation activities, whereas Tanah Bumbu is characterised by mining and rapid population growth driven by labour migration. Both districts are categorised as low endemic for malaria but maintain vulnerability due to occupational exposures and frequent mobility of adult populations, who are more likely to be affected in such settings. Study design and data collection This descriptive study used secondary data from March 2020 to May 2022. The study period was determined by: (1) the availability of Facebook Data for Good mobility data for Indonesia, which commenced in March 2020-May 2022 in response to the COVID-19 pandemic; and (2) the end of the Delta wave and initial recovery of malaria surveillance systems, which occurred in the first quarter of 2022. Malaria testing and case data were obtained from the SISMAL (Sistem Informasi Malaria), Indonesia’s national malaria surveillance system, managed by the Ministry of Health. SISMAL aggregated routine, case-based reports from health facilities, including public health centres, clinics, and hospitals, and captured information on confirmed cases, diagnostic methods, treatment, and geographic location. The system served as the primary platform for monitoring malaria trends, stratifying endemicity, and tracking progress toward malaria elimination at both district and national levels [4]. The positivity rate is calculated by confirmed cases divided by tests in each district during the study period. Endemicity classification of origin districts was based on the Annual Parasite Incidence (API) and data from the Ministry of Health during the study period, with category elimination (no indigenous case and API < 1‰), low endemic area (API < 1‰), moderate endemic area (API 1–5‰) and high endemic area (API > 5‰) [4]. District-level monthly COVID-19 case data were obtained from the routine surveillance system managed by District Health Offices (DHOs), which consolidated daily case reports from public health centres and hospitals within each district. Human mobility data were obtained from Facebook's Data for Good program [15], which provides aggregated, anonymised mobility statistics based on geotagged movements of Facebook users at the district level. This dataset, made public during the COVID-19 pandemic, enabled researchers to obtain near-real-time mobility estimates while maintaining strong user privacy and data security. Mobility estimates were generated using geotagged location data from Facebook users who had enabled location services. A mobility event is defined as origin–destination movements of a Facebook user between two district levels, based on observed user locations during specified time windows, not solely on posts or application activity. Facebook applies differential privacy techniques and aggregates movements to a minimum threshold to prevent re-identification. Therefore, these data reflect the movement patterns of Facebook users with location services enabled, not those of the general population. Movement into low-endemic districts poses a risk of malaria reintroduction, as individuals arriving from higher-transmission areas may import infections into receptive settings. Therefore, this study focused on incoming mobility to four targeted districts as a key pathway for potential case reintroduction and localised transmission. We acknowledge that outflow mobility, movements originating from the study districts toward other areas, was not analysed in this study. User IDs were replaced with anonymous geocodes, and time stamps were synchronized with the administrative boundaries of the study districts. Data processing and extraction were carried out using R software version 4.4.1, and algorithms were used to calculate mobility frequency rates for all district pairs. Data management All datasets used in this study were managed following data security and privacy protection principles. Malaria and COVID-19 surveillance data were obtained in de-identified form, with all personal identifiers removed prior to analysis. The datasets included only variables needed for epidemiological analysis, such as month, district, test and case counts, to minimize the risk of identifying individuals. Access to the data was limited to authorized members of the research team and stored on password-protected devices. Mobility data from Facebook were used only in aggregated form and did not contain any personally identifiable information, in line with Facebook’s privacy-preserving data protocols. All analyses and reporting were conducted using aggregated results to ensure that no individual patient or user could be identified. Data analysis Descriptive statistics, such as tables and graphs, were utilised to summarise the temporal and spatial trends of COVID-19 cases, malaria cases, and migration inflows. Multivariable Poisson regression models were used to investigate associations between mobility frequency, COVID-19 burden, malaria testing intensity, and malaria case counts, adjusting for district location and the endemicity of mobility origin. Two pre-specified multivariable models were compared using the Akaike Information Criterion (AIC) for goodness-of-fit: Model 1 included location, endemicity status of mobility origin categorized as elimination, low, moderate, and high endemic, mobility events per 1,000 population, malaria testing rate per 1,000 population, and COVID-19 case counts. Model 2 included the same covariates as Model 1 and additionally incorporated interaction terms between origin and destination endemicity. Model selection was based on AIC values together with the statistical significance of the interaction terms. Continuous variables were scaled per 1,000 units to improve the interpretability of regression coefficients without altering model estimates or inference. All estimates are interpreted as measures of association, and given the observational study design and analytical framework, causal inference cannot be established. Given the low-endemic nature of the study districts, zero monthly case counts were observed in some months (approximately 22% of Kulon Progo months). Standard Poisson regression was used as the primary analytical approach; however, we acknowledge that Zero-Inflated Poisson or Negative Binomial models may better accommodate overdispersion and zero inflation, and that sensitivity analyses using these models are recommended in future work. Results Overview of malaria testing and cases Across the study site, malaria testing and malaria confirmation cases were low relative to population size. Population size during the data period for Kulon Progo: 443,591 people; Purworejo: 787,257 people; Tanah Bumbu: 335,079 people; and Tanah Laut: 361,044 people [16–19]. In Java, Kulon Progo and Purworejo recorded similar volumes of malaria examinations, with 26,086 tests (46.6%) and 26,958 tests (48.2%), respectively (Table 1). Confirmed malaria cases were low in absolute terms relative to population size. Among all 872 confirmed cases across the four districts, Purworejo accounted for 647 (74.2%), Tanah Laut for 109 (12.5%), Tanah Bumbu for 50 (5.7%), and Kulon Progo for 66 (7.6%). In contrast, the districts outside Java conducted substantially fewer malaria tests, with 1,052 (1.9%) in Tanah Bumbu and 1,867 (3.3%) in Tanah Laut, yet exhibited higher positivity rates (4.75% and 5.83%, respectively) compared with districts in Java (Table 1). Table 1Description of the malaria testing, malaria cases, COVID-19 cases, and mobility data in four districts, March 2020–May 2022VariableJavaOutside JavaTotalKulon ProgoPurworejoTanah BumbuTanah Laut(n, %)(n, %)(n, %)(n, %)(n, %)Number of malaria tests26,086 (46.61)26,958 (48.17)1,052 (1.88)1,867 (3.34)55,963 (100)Number of malaria cases66 (7.57)647 (74.20)50 (5.73)109 (12.50)872 (100)Positivity rate0.25%2.40%4.75%5.83%1.56%Number of COVID-19 cases23,954 (39.22)21,308 (34.89)6,387 (10.46)9,426 (15.43)61,075 (100)Number of recorded mobility39,889 (8.30)114,266 (23.77)175,999 (36.62)150,421 (31.30)480,575 (100)Number of mobility by year March–Dec 202013,845 (34.71)41,462 (36.29)64,365 (36.57)54,938 (36.52)174,610 (36.33) January–Dec 202119,293 (48.37)53,916 (47.18)83,072 (47.20)71,008 (47.21)227,289 (47.30) January–May 20226,751 (16.92)18,888 (16.53)28,562 (16.23)24,475 (16.27)78,676 (16.37)Endemicity status of the location of origin mobility Elimination33,122 (83.04)100,417 (87.88)21,235 (12.06)77,487 (51.51)232,261 (48.33) Low endemic6,766 (16.96)13,849 (12.12)154,699 (87.89)72,933 (48.49)248,247 (51.65) Moderate endemic0 (0.00)0 (0.00)54 (0.03)1 (0.001)55 (0.01) High endemic1 (0.003)0 (0.00)11 (0.006)0 (0.00)12 (0.002)All percentages represent the proportion of each district's contribution to the total across all four study districts for the respective variable. Population size during the data period of Kulon Progo: 443,591 people; Purworejo: 787,257 people; Tanah Bumbu: 335,079; and Tanah Laut: 361,044 (BPS, 2022) Table 1Description of the malaria testing, malaria cases, COVID-19 cases, and mobility data in four districts, March 2020–May 2022VariableJavaOutside JavaTotalKulon ProgoPurworejoTanah BumbuTanah Laut(n, %)(n, %)(n, %)(n, %)(n, %)Number of malaria tests26,086 (46.61)26,958 (48.17)1,052 (1.88)1,867 (3.34)55,963 (100)Number of malaria cases66 (7.57)647 (74.20)50 (5.73)109 (12.50)872 (100)Positivity rate0.25%2.40%4.75%5.83%1.56%Number of COVID-19 cases23,954 (39.22)21,308 (34.89)6,387 (10.46)9,426 (15.43)61,075 (100)Number of recorded mobility39,889 (8.30)114,266 (23.77)175,999 (36.62)150,421 (31.30)480,575 (100)Number of mobility by year March–Dec 202013,845 (34.71)41,462 (36.29)64,365 (36.57)54,938 (36.52)174,610 (36.33) January–Dec 202119,293 (48.37)53,916 (47.18)83,072 (47.20)71,008 (47.21)227,289 (47.30) January–May 20226,751 (16.92)18,888 (16.53)28,562 (16.23)24,475 (16.27)78,676 (16.37)Endemicity status of the location of origin mobility Elimination33,122 (83.04)100,417 (87.88)21,235 (12.06)77,487 (51.51)232,261 (48.33) Low endemic6,766 (16.96)13,849 (12.12)154,699 (87.89)72,933 (48.49)248,247 (51.65) Moderate endemic0 (0.00)0 (0.00)54 (0.03)1 (0.001)55 (0.01) High endemic1 (0.003)0 (0.00)11 (0.006)0 (0.00)12 (0.002)All percentages represent the proportion of each district's contribution to the total across all four study districts for the respective variable. Population size during the data period of Kulon Progo: 443,591 people; Purworejo: 787,257 people; Tanah Bumbu: 335,079; and Tanah Laut: 361,044 (BPS, 2022) Description of the malaria testing, malaria cases, COVID-19 cases, and mobility data in four districts, March 2020–May 2022 All percentages represent the proportion of each district's contribution to the total across all four study districts for the respective variable. Population size during the data period of Kulon Progo: 443,591 people; Purworejo: 787,257 people; Tanah Bumbu: 335,079; and Tanah Laut: 361,044 (BPS, 2022) COVID-19 trends COVID-19 case notifications were concentrated in the Java districts, reflecting both higher population density and greater testing intensity. Kulon Progo reported 23,954 COVID-19 cases, accounting for 39.2% of all recorded cases, followed by Purworejo with 21,308 cases (34.9%). In contrast, Tanah Bumbu, the district outside Java, contributed 6,387 cases, representing 10.5% of the total (Table 1). Mobility patterns A total of 480,575 mobility events were documented during the study period, with the largest proportion of movements recorded in January–December 2021, accounting for approximately 47% of total mobility events (Table 1). Among the study sites, Tanah Bumbu had the highest mobility rate (36.6%), indicating a significant concentration of population movement in this district. Figure 1 illustrates the year-by-year transitions in mobility flows across malaria-endemicity levels (elimination, low, medium, and high) and their distribution across the four districts: Kulon Progo, Purworejo, Tanah Bumbu, and Tanah Laut. In terms of regional differences, distinct patterns emerged between Java and outside Java areas: (a) Java: most mobility events originated from elimination areas (> 80%), reflecting the dominance of malaria-free regions as sources of movement. (b) outside Java regions: mobility was primarily associated with low-endemic areas, accounting for 48.5% and 87.9% of total mobility events, highlighting the dominant role of these areas in shaping mobility dynamics outside Java. Movements from moderate- and high-endemic areas were minimal across all districts, indicating that the majority of population inflows originated from areas with low or no ongoing malaria transmission.Fig. 1The percentage of total incoming mobility events from each endemicity category across the four selected districts occurred between March 2020 and May 2022 The percentage of total incoming mobility events from each endemicity category across the four selected districts occurred between March 2020 and May 2022 Mobility, malaria testing and COVID-19 trends Mobility patterns were assessed in relation to trends in malaria cases (Fig. 2). In the Java Island (Fig. 2A), such as Purworejo, both mobility and malaria cases fluctuated, with malaria peaking in mid- to late 2021, briefly declining, and resurging in early 2022, indicating transmission was not yet controlled. In Kulon Progo, mobility remained relatively stable, with a slight increase in early 2021. Meanwhile, malaria cases fluctuated from late 2021, rising sharply in April and May 2022, indicating ongoing transmission risks despite stable mobility. These findings suggest that malaria transmission persisted in both districts, with case surges not fully associated with mobility, underscoring the complexity of transmission dynamics in Java. Outside Java Island (Fig. 2B), represented by Tanah Bumbu and Tanah Laut, mobility increased in early 2020 and then stabilized, exhibiting slight fluctuations. Malaria cases, however, were more volatile, with multiple peaks. Tanah Bumbu experienced surges in October 2020, March 2021, and February 2022, while Tanah Laut peaked in January 2021 and January 2022. Overall, no clear temporal link was observed between mobility and malaria trends in these districts, suggesting that other factors were more influential in driving transmission dynamics.Fig. 2Monthly mobility and malaria cases in: A Java Island (Purworejo and Kulon Progo) and B Outside Java Island (Tanah Bumbu and Tanah Laut). Each panel shows monthly malaria cases (left y-axis, absolute counts), and mobility events (right y-axis, number of Facebook mobility events per 1,000) displayed on independent scales to allow visual comparison of trends Monthly mobility and malaria cases in: A Java Island (Purworejo and Kulon Progo) and B Outside Java Island (Tanah Bumbu and Tanah Laut). Each panel shows monthly malaria cases (left y-axis, absolute counts), and mobility events (right y-axis, number of Facebook mobility events per 1,000) displayed on independent scales to allow visual comparison of trends We also examined differences in monthly malaria testing (compared with 2019 monthly data as a baseline) and COVID-19 cases during the study period (Fig. 3). Overall, malaria testing declined significantly at the onset of the COVID-19 pandemic, worsened during the Delta variant surge in mid-2021, and gradually recovered in 2022. In Java, Kulon Progo initially exhibited stable testing, followed by a sharp decline between December 2020 and July 2021, and then an inconsistent recovery (Fig. 3A). Purworejo experienced significant fluctuations, with a marked decrease in early 2020 and a gradual improvement by late 2021. Outside Java Island, Tanah Bumbu consistently maintained low but relatively stable testing levels, punctuated by spikes in July and October 2022 (Fig. 3B). By contrast, Tanah Laut experienced the steepest and least recovered decline, with only minimal rebound throughout the observation period. These patterns reflect pandemic-related disruptions in health system capacity, limiting access to routine malaria testing services.Fig. 3Monthly malaria testing and COVID-19 cases in: A Java Island (Purworejo and Kulon Progo) and B Outside Java Island (Tanah Bumbu and Tanah Laut) Monthly malaria testing and COVID-19 cases in: A Java Island (Purworejo and Kulon Progo) and B Outside Java Island (Tanah Bumbu and Tanah Laut) Regression analysis We compared two Poisson regression models and selected the optimal model using the Akaike Information Criterion (AIC) (Table 2). Model 1 was selected as the better-fitting model (AIC = 3,897.6 vs. Model 2 AIC = 3,899.2). In the multivariable Poisson regression (Model 1), mobility, malaria testing, and COVID-19 positivity were associated with malaria incidence [IRR 1.25 (95% CI 1.21–1.30), IRR 2.33 (95% CI 2.22–2.45), and IRR 1.12 (95% CI 1.08–1.17), respectively]. The IRR for malaria testing should be interpreted as reflecting surveillance effort (higher testing identifies more cases) rather than as a driver of transmission. The incidence of malaria outside Java was 64% lower than in Java (IRR = 0.36; 95% CI 0.29–0.44). Table 2Associations between malaria incidence and geographic location, endemicity of mobility origin, mobility events, malaria examination rates, and COVID-19 casesCharacteristicModel 1Model 2IRR95% CIp-valueIRR95% CIp-valueMobility events (n mobility events/1000)1.251.21, 1.30 < 0.0011.251.21, 1.30 < 0.001COVID-19 cases (n COVID-19 cases/1000)1.121.08, 1.17 < 0.0011.121.08, 1.17 < 0.001Malaria testing (n malaria testing/1000)2.332.22, 2.45 < 0.0012.332.22, 2.45 < 0.001Location Java–––– Outside Java0.350.29, 0.42 < 0.0010.360.30, 0.43 < 0.001Origin endemicity Elimination–––– Low endemic1.391.24, 1.56 < 0.0011.371.22, 1.55 < 0.001 High endemic0.560.25, 1.070.110.550.24, 1.050.1 Moderate2.411.41, 3.91 < 0.0011.971.04, 3.450.026Destination endemicity Low endemic–––– Elimination1.511.30, 1.75 < 0.0011.431.16, 1.75 < 0.001Interaction terms Origin endemicity x Destination endemicity Low endemic * Elimination––1.080.82, 1.410.6 High endemic * Elimination–– Moderate * Elimination––2.220.77, 5.750.11 Model parameter AIC38983899Model 1 included all five predictors (location, origin endemicity, destination endemicity, mobility events, malaria testing, and COVID-19 cases); Model 2 included all five predictors + interaction terms of origin endemicity and destination endemicity; Continuous variables increase per 1,000 unitsCI Confidence Interval, IRR Incidence Rate Ratio, AIC Akaike Information Criterion Table 2Associations between malaria incidence and geographic location, endemicity of mobility origin, mobility events, malaria examination rates, and COVID-19 casesCharacteristicModel 1Model 2IRR95% CIp-valueIRR95% CIp-valueMobility events (n mobility events/1000)1.251.21, 1.30 < 0.0011.251.21, 1.30 < 0.001COVID-19 cases (n COVID-19 cases/1000)1.121.08, 1.17 < 0.0011.121.08, 1.17 < 0.001Malaria testing (n malaria testing/1000)2.332.22, 2.45 < 0.0012.332.22, 2.45 < 0.001Location Java–––– Outside Java0.350.29, 0.42 < 0.0010.360.30, 0.43 < 0.001Origin endemicity Elimination–––– Low endemic1.391.24, 1.56 < 0.0011.371.22, 1.55 < 0.001 High endemic0.560.25, 1.070.110.550.24, 1.050.1 Moderate2.411.41, 3.91 < 0.0011.971.04, 3.450.026Destination endemicity Low endemic–––– Elimination1.511.30, 1.75 < 0.0011.431.16, 1.75 < 0.001Interaction terms Origin endemicity x Destination endemicity Low endemic * Elimination––1.080.82, 1.410.6 High endemic * Elimination–– Moderate * Elimination––2.220.77, 5.750.11 Model parameter AIC38983899Model 1 included all five predictors (location, origin endemicity, destination endemicity, mobility events, malaria testing, and COVID-19 cases); Model 2 included all five predictors + interaction terms of origin endemicity and destination endemicity; Continuous variables increase per 1,000 unitsCI Confidence Interval, IRR Incidence Rate Ratio, AIC Akaike Information Criterion Associations between malaria incidence and geographic location, endemicity of mobility origin, mobility events, malaria examination rates, and COVID-19 cases Model 1 included all five predictors (location, origin endemicity, destination endemicity, mobility events, malaria testing, and COVID-19 cases); Model 2 included all five predictors + interaction terms of origin endemicity and destination endemicity; Continuous variables increase per 1,000 units CI Confidence Interval, IRR Incidence Rate Ratio, AIC Akaike Information Criterion Overall, the analysis highlights that malaria resurgence during the COVID-19 pandemic is multifactorial, associated with geographic location, mobility dynamics, testing intensity, and concurrent COVID-19 transmission, rather than mobility alone. Discussion The findings of this study demonstrate that malaria incidence during the COVID-19 pandemic was associated with multiple factors, including district location, the origin and intensity of population mobility events, malaria diagnostic testing intensity, and COVID-19 burden. These associations suggest, but do not establish causally, that malaria patterns in low-endemic settings during the pandemic reflected a multifactorial process. Importantly, the apparent resurgence in case counts observed in late 2021–2022 may partly reflect improved testing recovery and delayed case identification rather than a true increase in transmission, given the known underreporting during earlier pandemic phases. This underscores that malaria resurgence in low-endemic settings during the pandemic cannot be associated with mobility alone but reflects a multifactorial process involving epidemiological context and health system performance. Disruptions to malaria surveillance, diagnosis, and treatment services during COVID-19 have been widely documented and are known to compromise timely case detection and transmission interruption [20, 21]. Against this backdrop, this study makes three key contributions: (1) it leveraging near-real-time digital mobility data (Facebook Data for Good) to assess mobility–malaria associations across multiple low-endemic districts in Indonesia; (2) examining these associations within the unique context of pandemic-related mobility disruptions; and (3) providing an empirical basis for integrating digital mobility indicators into Indonesia’s national malaria surveillance system. Our study demonstrated substantial disruptions to malaria testing in both Java and outside Java districts during the COVID-19 pandemic, particularly during the initial phase in 2020 and the surge of the Delta variant in mid-2021. These patterns are consistent with global evidence showing that COVID-19 strained health systems and diverted resources away from routine services, including malaria surveillance, diagnosis, and case management [8, 22]. Reduced malaria testing likely delayed case detection, increasing the risk of undetected transmission, particularly in low-endemicity and elimination settings. Although a gradual recovery in testing was observed in 2022, levels had not fully returned to pre-pandemic capacity, suggesting lingering impacts on diagnostic services. Similar modelling studies have warned that even short-term interruptions in malaria diagnosis can be associated with disproportionate increases in transmission and mortality [22, 23]. District-level variation in malaria testing and recovery trajectories underscores the critical role of health system resilience during overlapping public health crises. Districts with stronger baseline surveillance capacity and more flexible workforce arrangements were better able to maintain essential malaria services despite COVID-19 surges. At the same time, those with limited diagnostic infrastructure or heavy pandemic burdens experienced prolonged disruptions. In Java, fluctuations in malaria testing in elimination-phase districts such as Purworejo and Kulon Progo highlight how even short-term surveillance lapses can precipitate outbreaks in low-transmission settings. Outside Java, the pronounced and incomplete recovery observed in Tanah Laut aligns with national bottleneck analyses identifying diagnostic shortages, workforce reallocation, and weakened reporting systems in parts of Kalimantan [20]. Similar patterns have been documented globally, with health systems with limited surge capacity struggling to sustain malaria surveillance amid competing COVID-19 demands [10]. Reports from the World Health Organization further emphasize that resilience, defined as the ability to absorb shocks while maintaining core functions, was a key determinant of malaria service continuity during and after the pandemic [21]. Strengthening resilient surveillance systems, protecting diagnostic capacity, and integrating outbreak preparedness into routine malaria programs are therefore essential to safeguard gains toward elimination against future health emergencies. In Java, cases of mobility and malaria demonstrated fluctuating patterns. Notably, malaria outbreaks were reported in Purworejo during 2021–2022 and in Kulon Progo in 2022, highlighting local vulnerabilities despite overall progress in elimination [14]. The mobility-malaria relationship in Java is linked by high-density social interaction in a near-elimination setting, where frequent population movement across densely populated districts amplifies the consequences of even sporadic importation. It also illustrates that an elimination status does not guarantee zero risk in the presence of gaps or imported cases [8]. These outbreaks suggest that, although most mobility originates from elimination areas, residual transmission risk persists due to gaps in surveillance, delayed outbreak detection, and challenges in sustaining elimination gains. Conversely, outside Java, there was no clear relationship between mobility patterns and trends in malaria cases. Outside Java, the driver is qualitatively different: Tanah Bumbu and Tanah Laut are characterised by mobile migrant populations (miners, plantation workers, and forest-adjacent labourers) whose movements are occupationally structured and ecologically targeted, repeatedly bringing individuals into forested high-transmission micro-environments. In the low-endemic study districts, the primary pathway through which mobility is hypothesised to affect malaria incidence is case importation, individuals arriving from higher-endemic areas carrying Plasmodium infections into receptive, low-immunity settings, rather than changed contact with local vector habitats. This distinction is important for interpreting our findings and for tailoring surveillance responses. Reported malaria cases in these regions were often associated with specific high-risk groups, including mobile or migrant populations such as forest goers, plantation workers, and miners. These groups typically move into or within high-transmission zones, contributing disproportionately to case burdens, while their mobility is often underrepresented in routine surveillance data [3]. A recent case–control study in Aceh found that individuals who sleep or live in forest settings or have secondary forest residences are at substantially higher odds of contracting malaria, suggesting that these mobile or semi-permanent occupational or ecological exposures are critical drivers of transmission in non-elimination areas [24]. Together, these sources underscore that Java’s risk is associated with mobility from elimination settings and occasional resurgences, whereas outside Java transmission is more closely linked to mobile, high-risk subgroups. Interventions must thus differ from intense surveillance and outbreak response in Java, and targeted strategies (active case finding, protective measures, and outreach) among mobile, forest-exposed populations outside Java. An important consideration when interpreting the contemporaneous monthly associations in this study is the temporal lag between a mobility event and its manifestation as a reported clinical case. This lag comprises at least three sequential components. First, the biological incubation period: for Plasmodium vivax, the dominant species in the Java and South Kalimantan districts, the incubation period ranges from 12 to 17 days, with dormant hypnozoites capable of causing relapse for months or years after primary infection. This means that a mobility event in a given month may not generate a detectable case until the following month or later. Second, health-seeking delays were compounded during the COVID-19 pandemic, as reduced facility attendance, overlapping fever symptoms, and staff reallocation to COVID-19 response introduced additional delays between symptom onset and case detection. Third, surveillance reporting lags, particularly in districts outside Java with more limited health infrastructure, further displaced the temporal signal between mobility and reported cases. These layered delays, estimated at 2–8 weeks in analogous near-elimination settings [6, 25, 26], likely attenuated the contemporaneous associations observed in our monthly ecological analysis, particularly in Tanah Bumbu and Tanah Laut, where no clear temporal pattern was identified. In this study, Facebook mobility indicators offered a practical proxy for monitoring movement into low-endemic malaria districts, a key pathway for potential case importation. Facebook Data for Good only provides re-aggregated data at certain administrative levels due to privacy concerns [27]. However, the dataset has important limitations. Facebook Data for Good captures only users who have enabled location services, creating a systematic demographic bias toward younger, urban, higher-income, and smartphone-owning users. Populations most likely to import malaria, mobile workers from Papua, NTT, and Kalimantan, including miners, forest workers, and seasonal migrants, are precisely those least likely to be captured in these data. Indonesia is one of the world's largest Facebook markets, with approximately 120 million users; however, platform penetration varies across age groups, socioeconomic strata, and geographic regions [28]. This likely explains the near-absence of mobility events from high-endemic origin areas (n = 12 records, 0.002% of total), which should not be interpreted as evidence of negligible high-risk movement but rather as a data capture limitation. The IRR estimate for mobility from high-endemic areas (IRR = 0.56) is therefore unreliable due to extreme data sparsity and should not be interpreted substantively. Although the dataset has limitations, such as dependence on users who have enabled location services, non-representativeness of certain demographic groups (e.g., older adults, lower-income rural populations, and high-risk occupational groups such as miners and forest workers), and variable spatial accuracy, it is well recognised for its utility in real-time mobility surveillance. The substantial heterogeneity in ecological setting, population characteristics, and health system capacity across the four study districts warrants caution in interpreting pooled regression results. The pooled IRR for mobility (1.25) represents an average association across districts with markedly different transmission contexts. Stratified analyses or district-specific models, informed by site-level epidemiological theory, would better characterise the heterogeneous nature of the mobility-malaria relationship across settings. Furthermore, all findings in this study are associational: the analytical approach used multivariable Poisson regression with AIC-based model comparison, which is appropriate for exploratory associational analysis in this data context but does not permit causal inference [6, 9, 29]. Our findings are consistent with the global literature demonstrating that human mobility is a primary driver of malaria importation into low-endemic and elimination-phase settings [29, 30]. Within Indonesia, the findings align with the Ministry of Health's recognition of imported malaria as a key threat to the 2030 elimination goal, as articulated in the 2025–2045 Malaria Elimination Roadmap [14]. The use of digital mobility data for malaria surveillance represents an emerging methodological approach with demonstrated utility in Zanzibar [6] and China [9]. Study limitations This study has several limitations. First, the Facebook mobility data are subject to demographic selection bias, systematically underrepresenting older adults, lower-income populations, and high-risk occupational groups (miners, forest workers, seasonal migrants) most relevant to malaria importation. Mobility estimates from Facebook should therefore be interpreted cautiously, but they remain a useful complementary tool alongside routine surveillance systems [31]. Second, malaria surveillance data from SISMAL are subject to underreporting, particularly during the COVID-19 pandemic phases. Third, the analysis assessed contemporaneous monthly associations without incorporating lag structures; the composite biological-behavioural-reporting lag of 2–8 weeks likely attenuated true associations. Fourth, standard Poisson regression was used without formal testing for overdispersion or zero-inflation, which may affect standard error estimates. Fifth, district-level aggregated analysis cannot account for within-district heterogeneity in transmission risk. Sixth, environmental covariates (rainfall, temperature) and intervention disruption data (LLIN distribution, IRS coverage) were not included due to data limitations. Seventh, all findings are associational and do not permit causal inference. Eighth, the four districts were purposively selected and may not represent all low-endemic settings in Indonesia. Finally, the mobility dataset captures only inflows; incorporating outflow mobility in future analyses would provide a more comprehensive understanding of transmission dynamics. As both inflow and outflow movements can differentially influence malaria spread, their joint assessment using appropriate data is recommended. Despite the limitation, we are confident that the demographic profile of Facebook users in Indonesia supports representing the demographic profile of inter-island movement in Indonesia. Approximately 64% of Facebook users are aged 18–34 years [28], closely matching the cohort with the highest propensity for inter-island movement in Indonesia, working-age individuals aged 15–34 years [32, 33]. Groups underrepresented on the platform, older adults and those with limited internet access are also the least mobile, meaning their exclusion introduces minimal distortion to the estimates. The demographic limitations of Facebook-derived data are therefore unlikely to compromise the validity of the study findings. Future research should adopt causal inference frameworks, including directed acyclic graphs (DAGs) for principled covariate selection, and use population-denominated incidence rates to improve comparability across settings. The Distributed Lag Non-linear Model (DLNM) at weekly resolution is a methodological priority, as it can simultaneously capture the lagged and non-linear effects of population mobility on malaria risk, relationships that are epidemiologically plausible but poorly characterised by conventional approaches. Applying DLNM in elimination settings would provide a more rigorous basis for understanding mobility-driven importation and informing targeted malaria control strategies. Conclusions Mobility remains a crucial factor in malaria control, although its relationship is complex and context-dependent, influenced by local endemicity and health system capacity. The COVID-19 pandemic demonstrated how mobility, service disruptions, and ineffective surveillance interact to affect disease transmission. These findings are associated in nature; causal inference requires explicit causal modelling approaches, not applied in this study. In areas nearing elimination, fluctuations in mobility and inadequate monitoring may be linked to outbreaks if imported cases are not quickly detected. In more endemic settings, transmission was linked to high-risk mobile groups, including forest dwellers and miners, and may be worsened by gaps in diagnostic access. Targeted interventions during such periods are essential to prevent a resurgence of the disease. Specifically, the National Malaria Control Program should consider integrating real-time digital mobility indicators, such as Meta Data for Good movement data, as a supplement to SISMAL to generate mobility-informed risk alerts for districts experiencing elevated inflows from endemic areas. This would enable proactive screening of high-mobility travellers, rapid deployment of diagnostic and case management resources during periods of increased importation risk, consistent with Indonesia's 2025–2045 Malaria Elimination Roadmap goals. Supplementary Information Supplementary material 1: Figure S1. Indonesia map by the district endemicity and four selected study sites. Kulon Progo and Purworejo districts represent the Java study site; Tanah Laut and Tanah Bumbu districts represent the outside Java study site. Supplementary material 1: Figure S1. Indonesia map by the district endemicity and four selected study sites. Kulon Progo and Purworejo districts represent the Java study site; Tanah Laut and Tanah Bumbu districts represent the outside Java study site.