This study is the first to our knowledge in the Philippines context to identify a pregnant woman with severe malaria who was admitted to the largest government referral hospital in the province and was located 230 km from her residence. Documentation of this condition during her pregnancy would not have been matched deterministically nor followed through by the local health authorities of Rizal and Palawan since both records had multiple errors in their names. Also, this study emphasizes the importance of integrating RWD with surveillance data to monitor events during the course of pregnancy which otherwise would remain unknown when reviewed for assessing disease program impact. As a result, the case might have been counted as an uncomplicated MDP case with full-term delivery to a live baby girl at 38 weeks of gestation, while her hospital records reveal that she had severe malaria with anemia and septic shock. Missing such medical information as comorbidities could bias the results of retrospective studies, especially on pregnancy [27, 28].
In the current Philippine health system, malaria surveillance and hospital administrative data are assessed separately and are not adequately utilized or hybridized for research such as epidemiology studies. However, through the integration of two databases, we gained a clearer understanding of true-to-life conditions among vulnerable populations in remote areas that otherwise would not be known. The true-positive matched patient was infected with malaria during the second and third trimester of her pregnancy. On the one hand, malaria surveillance data does not include detailed information about a woman’s pregnancy such as the last menstrual period. On the other hand, hospital data lacks information about other endemic diseases in the community such as malaria unless these are suspected in a pregnant woman and tests are performed. This highlights the importance of integrated information. Obtaining such information about the period of infection among these MDP patients, as well as the duration of infection and treatment, is important for investigating the accurate relationship between the malaria infection and pregnancy outcome. Record linkage allows us to retrospectively estimate these conditions as well as first-trimester malaria infections, when most of these women could have been reported as non-pregnant at the first point-of-contact at the barangay health station. Resolving such concerns about the missing data of the last menstrual period that are often excluded in studies assessing anti-malaria treatment in pregnancy can possibly achieve a more robust result [18, 19, 23]. Therefore, integration and usage of RWD have the potential to provide information that allows a better assessment of the impact disease programs on vulnerable populations. This underlies the Philippines DOH Administrative Order No. 2009-0025 outlining policies and principles of MCH implementation of hospital delivery and attendance by skilled staff, along with initiatives for newborn and child healthcare from the Administrative Order No. 2008-0029 regarding Health Reforms for Rapid Reduction of Maternal and Newborn Mortality.
Despite national concerns and expectations for health prevention in a patient's life course approach, especially among pregnant women, there remain many gaps among multiple databases due to independent collection and inconsistency in recording. Thus, patient identification from multiple sources becomes difficult, and the utilization of data for assessing the disease becomes complicated. Even though a pregnant woman with a history of fever will be tested for malaria, this event will be captured in the malaria surveillance only if she sought consultation at the rural clinic or disclosed her fever history during her prenatal check-ups. Further, a pregnant woman with malaria could be asymptomatic during her first prenatal checkup. The matched case in this study might have been taken to ONP because of symptoms of severe anemia and septic shock. We later discovered from malaria surveillance records that she was also poorly nourished which could affect fetal growth. To fill in these gaps, universal health coverage and data harmonization such as applying unique patient identifiers with a uniformed method for health checkup registries, newborn registries, surveillance, and healthcare claims data, are important for better comprehension of the disease and for improving clinical outcomes. Thus, probabilistic record linkage could play such a role for underutilized data.
When looking closely at the background characteristics of the mutual variables used for matching, there was a significant demographic difference between pregnant women admitted to ONP and MDP patients with regard to string lengths of full name, age, and village of residence. These differences may have reflected the result of matching scores resulting in only one true-positive match. Also, from the non-MDP patient’s characteristics at the hospital, about a third were primigravida; further, 10% were diagnosed with preeclampsia. Although outcomes for newborns were not obtained for non-matched patients, the mean age between MDP and non-MDP patients is significantly different and needs to be adjusted when assessing maternal and newborn outcomes. With age as an example, propensity score matching (PSM) can be applied to observational studies to reduce biases when examining the impact of exposure or intervention. This is done thru adjusting similar propensity scores such as patient characteristics, co-morbidities, disease severity and treatment to ensure a balanced distribution of observed covariates between the two groups [29, 30]. However, studies using this method are limited to malaria and should be applied when using RWD for observing the associated risks [31]. Although we followed the course of pregnancy and delivery within the true-positive matched case, important variables such as drug treatment and duration could not be investigated unless patient charts were reviewed. These factors can be included in the analysis after integrating maternal clinical information from medical charts and by increasing the number health facilities in the review.
Considering the probabilistic record linkage, the proportion of true-positive patients exceeded when the 70% threshold was set for fuzzy matching in this study. In previous studies, 80% is said to be used for assuring true-positivity [11, 32]. This was probably affected due to the majority of false-positive patients which was around 36.7 compared to only one true-positive patient with a score of 97.7. Although we could not find any false negative matches in this study, false-positive patients could overlap in between the thresholds, and therefore, validation within the overlapping patient’s scores is necessary as more true-positive matches increase [33]. Alongside that, applying independent weights to missing data and prioritizing the variables depending on likelihood of matching can lead to a more robust result for reliability and accuracy of true-positive matches [21, 34]. Calculation of ROC curves and AUC scores are necessary when assessing the precision and recall of the fuzzy matching [35]. This study, however, contained only one true-positive match and therefore we could not investigate the aforementioned approach. Incorporating these aspects for validation are necessary in the future development of the database.
Though this record linkage is useful for utilizing multiple databases, there are still many obstacles to developing a comprehensive real-world database for epidemiological studies. We limited our search to patients admitted to the largest referral hospital over 200km away from the highly endemic malaria area and only one patient was considered a true-positive match. This means that the majority of the MDP patients registered in the surveillance system remain unidentified; therefore, our results cannot be generalized. In addition, other MDP patients are likely to have delivered their babies in health facilities closer to home. Although facility-based delivery (FBD) increased after the enhancement of the PhilHealth insurance coverage and the implementation of the ‘no home-birthing policy [36], substantial proportion of home deliveries still occur especially in rural areas of the Philippines [37, 38]. It is reported from a previous study that barangays with monetary sanction are associated with higher FBD, but home deliveries remain due to several factors such as affordability, accessibility, and tradition [39, 40, 41]. Therefore, integration of maternal and newborn clinical records in rural hospitals, as well as newborn records of home deliveries need is essential within these groups for association with malaria or any disease in future pregnancy outcomes.