Probabilistic Record Linkage Research Articles

Implicates as Instrumental Variables: An Approach for Estimation and Inference with Probabilistically Matched Data

Abstract Linkage errors in probabilistically matched data sets can cause biases in the estimation of regression coefficients. This article proposes an approach to obtain consistent estimates and valid inference that relies on instrumental variables. The novelty of the method is to show that instrumental variables arise naturally in the course of probabilistic record linkage thereby allowing for off-the-shelf implementation. Relative to existing approaches, the instrumental variable approach does not require integration of the record linkage and regression analysis steps, the estimation of complex models of linkage error, or computationally expensive methods to estimate standard errors. The instrumental variables approach performs well in Monte Carlo simulations of an environment highlighting a many-to-one linkage problem.

Linking population-based cohorts with cancer registries in LMIC: a case study and lessons learnt in India

ObjectivesIn resource-constrained settings, cancer epidemiology research typically relies on self-reported diagnoses. To test a more systematic alternative approach, we assessed the feasibility of linking a cohort with a cancer registry.SettingData...

Perinatal outcomes and genomic characteristics of fetal copy number variants: An individual record linkage study of 713 pregnancies.

To determine the perinatal outcomes of fetuses diagnosed with a pathogenic copy number variant (CNV) or variant of uncertain significance (VUS); and to characterize these variants in terms of testing indication, genomic location, size, and inheritance. Retrospective study of singleton pregnancies with a pathogenic CNV or VUS from a single laboratory during 2012-2018. Probabilistic record linkage between the prenatal diagnosis dataset and perinatal outcome data for births from 20weeks gestation was performed. If no birth record was found, this implied a pregnancy loss <20weeks. We included 6945 prenatal microarray results; a pathogenic CNV was detected in 230 (3.3%, 95% CI: 2.9%-3.8%) and a VUS in 483 (7.0%, 95% CI: 6.4%-7.6%). Of pregnancies with a pathogenic CNV, 20.0% (95% CI: 15.3%-25.6%) had a live birth, 3.0% (95% CI: 1.5%-6.2%) had a perinatal death (stillbirth or neonatal death), and 77% (95% CI: 71.1%-81.9%) had no birth record. Of those with a VUS, 64.4% (95% CI: 60.0%-68.5%) had a live birth, 1.8% (95% CI: 1.0%-3.5%) had a perinatal death, and no birth record was found for 33.7% (95% CI: 29.7%-38.1%). Most pathogenic CNVs (61.1%) were <7Mb in size. The most common microdeletion syndromes were DiGeorge, Wolf-Hirschhorn, and Cri-du-chat syndromes. This study provides an overview of perinatal outcomes and frequency of recurrent CNVs observed in the prenatal microarray era.

Validating matches of electronically reported fishing trips to investigate matching error

AbstractObjectiveRecently, researchers have developed methods for combining probability samples and non‐probability samples. In recreational fisheries management, data from probability samples are typically counts of catch from a random sample of trips intercepted by a sampler, while non‐probability samples consist of catch data that are collected in self‐reports made to a fishery management agency. These reports are typically transmitted electronically and are known as an electronic logbook (ELB). Even when such reporting is mandated, compliance is not universal. Since the inclusion probability for any particular angler is unknown, the ELB sample is a non‐probability sample. We used data from a 2017 Gulf of Mexico (GoM) pilot study in which charter captains volunteered to electronically report their catch. At the dock, they could also be intercepted by a sampler, at which time their catch was observed. Estimates of total catch can be generated if trips from the two data sets can be accurately matched. Several states in the GoM implement similar ELB reporting augmented with a probability sample. However, there is an apparent discrepancy between National Oceanic and Atmospheric Administration (NOAA) estimates of the total and the ELB estimates for the same geographies. We seek to investigate the extent to which matching errors contribute to the discrepancies.MethodsWe employed probabilistic record linkage to match reports with intercepts and developed a validation tool to examine the matches. Using our validation tool, we examined several methods of estimating the total catch of Red Snapper Lutjanus campechanus in the GoM to investigate the potential cause of the discrepancy.ResultWe found the existing differences between the NOAA estimates and estimates resulting from combining ELB reports in this application were likely not due to matching error but instead were apparently derived from other sources of non‐sampling error.ConclusionThis has implications for new and existing ELB implementations, which are gaining popularity. The tool and results we provide can allow other implementations to better match reports with intercepts and offers a way to examine the extent to which matching errors affect the bias of estimates. Our work also shows that agencies should focus on non‐sampling errors besides matching error to reduce bias. Our tool may also be extended to examine such non‐sampling errors, such as the assumption that reporting and interception are independent.

Improving Probabilistic Record Linkage Using Statistical Prediction Models

SummaryRecord linkage brings together information from records in two or more data sources that are believed to belong to the same statistical unit based on a common set of matching variables. Matching variables, however, can appear with errors and variations and the challenge is to link statistical units that are subject to error. We provide an overview of record linkage techniques and specifically investigate the classic Fellegi and Sunter probabilistic record linkage framework to assess whether the decision rule for classifying pairs into sets of matches and non‐matches can be improved by incorporating a statistical prediction model. We also study whether the enhanced linkage rule can provide better results in terms of preserving associations between variables in the linked data file that are not used in the matching procedure. A simulation study and an application based on real data are used to evaluate the methods.

Overall survival analyses of female malignancies in Southern Brazil during 2008-2017: A closer look at breast, cervical and ovarian cancer.

The aim of this study was to report the overall survival and baseline factors associated with OS for breast, cervical and ovarian cancer in Florianópolis, Southern Brazil, a region with quality-of-life indicators comparable to high-income countries. Cohort study was performed from probabilistic record linkage of the Mortality Information System and the Population-based cancer registry of Florianópolis. It was included breasts, cervical and ovarian cancer diagnosis during the period of 2008-2012 with a follow up of 60 months. Cox regression and Kaplan-Meier method were used for associations with overall survival and risk factors. 1857 cases of the three malignancies were included in the analysis. We identified 202 deaths in breast cancer subjects, 53 for cervical cancer and 51 for ovarian cancer. Metastatic disease at diagnosis was present in 31%, 9.6%, and 55% of the cases, respectively. Overall survival was statistically correlated with age, educational level and stage for breast cancer; age and stage for cervical cancer; age and stage for ovarian cancer. Metastatic disease and age are the main prognostic factors for the malignancies studied, as they were associated with both overall survival and risk of death. Better screening and preventive tests for early diagnosis are needed. Support of Research and Innovation in the State of Santa Catarina, Research Program for the Unified Health System (FAPESC/MS-DECIT/CNPQ/SES-SC-PPSUS); the Brazilian National Research Council (CNPq); and the Coordination for the Improvement of Higher Education Personnel (CAPES).

Validating a novel deterministic privacy-preserving record linkage between administrative & clinical data: applications in stroke research.

Research data combined with administrative data provides a robust resource capable of answering unique research questions. However, in cases where personal health data are encrypted, due to ethics requirements or institutional restrictions, traditional methods of deterministic and probabilistic record linkages are not feasible. Instead, privacy-preserving record linkages must be used to protect patients' personal data during data linkage. To determine the feasibility and validity of a deterministic privacy preserving data linkage protocol using homomorphically encrypted data. Feasibility was measured by the number of records that successfully matched via direct identifiers. Validity was measured by the number of records that matched with multiple indirect identifiers. The threshold for feasibility and validity were both set at 95%. The datasets shared a single, direct identifier (health card number) and multiple indirect identifiers (sex and date of birth). Direct identifiers were encrypted in both datasets and then transferred to a third-party server capable of linking the encrypted identifiers without decrypting individual records. Once linked, the study team used indirect identifiers to verify the accuracy of the linkage in the final dataset. With a combination of manual and automated data transfer in a sample of 8,128 individuals, the privacy-preserving data linkage took 36 days to match to a population sample of over 3.2 million records. 99.9% of the records were successfully matched with direct identifiers, and 99.8% successfully matched with multiple indirect identifiers. We deemed the linkage both feasible and valid. As combining administrative and research data becomes increasingly common, it is imperative to understand options for linking data when direct linkage is not feasible. The current linkage process ensured the privacy and security of patient data and improved data quality. While the initial implementations required significant computational and human resources, increased automation keeps the requirements within feasible bounds.

Extending the Fellegi-Sunter record linkage model for mixed-type data with application to the French national health data system

Probabilistic record linkage is a process of combining data from different sources, when such data refer to common entities and identifying information is not available. A probabilistic record linkage framework that takes into account multiple non-identifying information that this is limited to simple binary comparison between matching variables has been previously proposed. An extension of this method is proposed for mixed-type comparison vectors. A mixture model for handling comparison values of low prevalence categorical matching variables, and a mixture of hurdle gamma distribution for handling comparison values of continuous matching variables have been developed. The parameters are estimated by means of the Expectation Conditional Maximization (ECM) algorithm. Through a Monte Carlo simulation study, both the posterior probability estimation for a record pair to be a match and the prediction of matched record pairs are evaluated. The simulation results indicate that the proposed methods outperform existing ones in most considered cases. The proposed methods are applied on a real dataset, to perform linkage between a registry of patients suffering from venous thromboembolism in the Brest district area (GETBO) and the French national health information system (SNDS).

The Data-Adaptive Fellegi-Sunter Model for Probabilistic Record Linkage: Algorithm Development and Validation for Incorporating Missing Data and Field Selection.

BackgroundQuality patient care requires comprehensive health care data from a broad set of sources. However, missing data in medical records and matching field selection are 2 real-world challenges in patient-record linkage.ObjectiveIn this study, we aimed to evaluate the extent to which incorporating the missing at random (MAR)–assumption in the Fellegi-Sunter model and using data-driven selected fields improve patient-matching accuracy using real-world use cases.MethodsWe adapted the Fellegi-Sunter model to accommodate missing data using the MAR assumption and compared the adaptation to the common strategy of treating missing values as disagreement with matching fields specified by experts or selected by data-driven methods. We used 4 use cases, each containing a random sample of record pairs with match statuses ascertained by manual reviews. Use cases included health information exchange (HIE) record deduplication, linkage of public health registry records to HIE, linkage of Social Security Death Master File records to HIE, and deduplication of newborn screening records, which represent real-world clinical and public health scenarios. Matching performance was evaluated using the sensitivity, specificity, positive predictive value, negative predictive value, and F1-score.ResultsIncorporating the MAR assumption in the Fellegi-Sunter model maintained or improved F1-scores, regardless of whether matching fields were expert-specified or selected by data-driven methods. Combining the MAR assumption and data-driven fields optimized the F1-scores in the 4 use cases.ConclusionsMAR is a reasonable assumption in real-world record linkage applications: it maintains or improves F1-scores regardless of whether matching fields are expert-specified or data-driven. Data-driven selection of fields coupled with MAR achieves the best overall performance, which can be especially useful in privacy-preserving record linkage.

Linking the Center for International Blood and Marrow Transplant Research Registry to the California Cancer Registry and California Hospital Patient Discharge Data

Advances in hematopoietic cell transplantation (HCT) have substantially improved patient survival, increasing the importance of studying outcomes and long-term adverse effects in the rapidly growing population of HCT survivors. Large-scale registry data from the Center for International Blood and Marrow Transplant Research (CIBMTR) are a valuable resource for studying mortality and late effects after HCT, providing detailed data reported by HCT centers on transplantation-related factors and key outcomes. This study was conducted to evaluate the robustness of CIBMTR outcome data and assess health-related outcomes and healthcare utilization among HCT recipients. We linked data from the CIBMTR for California residents with data from the population-based California Cancer Registry (CCR) and hospitalization information from the California Patient Discharge Database (PDD). In this retrospective cohort study, probabilistic and deterministic record linkage used key patient identifiers, such as Social Security number, ZIP code, sex, birth date, hematologic malignancy type and diagnosis date, and HCT type and date. Among 22,733 patients registered with the CIBMTR who underwent autologous or allogeneic HCT for hematologic malignancy between 1991 and 2016, 89.0% were matched to the CCR and/or PDD (n=17,707 [77.9%] for both, n=1179 [5.2%] for the CCR only, and n=1342 [5.9%] for the PDD only). Unmatched patients were slightly more likely to have undergone a first autologous HCT than an allogeneic HCT (12.6% versus 9.0%), to have a larger number of missing linkage identifiers, and to have undergone HCT prior to 2010. Among the patients reported to the CIBMTR who matched to the CCR, 85.7% demonstrated concordance of both hematologic malignancy type and diagnosis date across data sources. This linkage presents unparalleled opportunities to advance our understanding of HCT practices and patient outcomes.

Score Test for Assessing the Conditional Dependence in Latent Class Models and its Application to Record Linkage

AbstractThe Fellegi–Sunter model has been widely used in probabilistic record linkage despite its often invalid conditional independence assumption. Prior research has demonstrated that conditional dependence latent class models yield improved match performance when using the correct conditional dependence structure. With a misspecified conditional dependence structure, these models can yield worse performance. It is, therefore, critically important to correctly identify the conditional dependence structure. Existing methods for identifying the conditional dependence structure include the correlation residual plot, the log-odds ratio check, and the bivariate residual, all of which have been shown to perform inadequately. Bootstrap bivariate residual approach and score test have also been proposed and found to have better performance, with the score test having greater power and lower computational burden. In this paper, we extend the score-test-based approach to account for different conditional dependence structures. Through a simulation study, we develop practical recommendations on the utilisation of the score test and assess the match performance with conditional dependence identified by the proposed method. Performance of the proposed method is further evaluated using a real-world record linkage example. Findings show that the proposed method leads to improved matching accuracy relative to the Fellegi–Sunter model.

Splink: Free software for probabilistic record linkage at scale.

Funded by ADR UK, a new data linking team at the Ministry of Justice set out to link administrative datasets across the justice space, for internal use and sharing with external researchers. To achieve this aim we sought a linkage implementation that was probabilistic, flexible, scalable and ideally open source. Taking into account the tools available at the MoJ, existing open-source software (and paid alternatives) failed to meet our desired criteria. It was decided to develop a software package that builds on FastLink’s implementation in R of an Expectation-Maximisation algorithm to estimate a Fellegi-Sunter linkage model, adding a range of technical improvements, increased functionality and customisation options. Distributed computing offered by Spark could facilitate comparable linkage jobs that run on much larger datasets and much faster. Working with government data, accountability and transparency are vital, so the data and models are made accessible by a range of intuitive visualizations. The Splink python package has been downloaded 2 million times. This initially used Spark to deliver its superior performance, but Splink v3 caters for various SQL backends and more potential users. Splink and supplementary python libraries are publicly visible on GitHub and provide assistance in all aspects of data linkage: - splink_data_standardisation - functions to perform general data standardisation - splink_cluster_studio - creates an interactive HTML dashboard to analyse record clusters (building on splink_graph - a library for generating graph metrics) - splink_synthetic_data - generating realistic synthetic data for testing linkage algorithms - splink_demos - interactive demo/tutorial notebooks for a range of features of the Splink libraries These tools are in continuous development and have already been used to deliver deduplicated and linked data products for the entire criminal justice system. Through technical innovation and user-focused development, Splink has improved access to cutting-edge data linkage, and created groundbreaking research opportunities at MoJ and beyond. The team is grateful to ONS and other collaborators for testing and adopting these tools, and will continue to explore ways to improve performance and user experience.

Piloting and validating the linkage of routine ART and laboratory records in an HIV endemic setting.

ObjectivesIntegrating clinical and laboratory data systems offers opportunities to understand disease aetiology and improve patient outcomes. We developed and piloted a data linkage algorithm to link HIV patient treatment records with routine laboratory service records in an HIV endemic setting in rural South Africa. ApproachWe used data from the South Africa’s National Health Laboratory Service (NHLS) and Three Integrated Electronic Registers (TIER.Net) databases for 17 primary healthcare clinics from uMkhanyakude district, South Africa. The two databases contained data on key demographic variables such as patient first name, surname, date of birth, sex, and health facility, and national ID numbers for a subset of the data. We adapted a probabilistic record linkage algorithm we previously developed for linkage of laboratory results. We trained and validated our algorithm using two approaches: 1) firstly, we constructed a quasi-gold standard based on manual review of potential matches for 1,061 randomly selected patients; 2) secondly, we constructed a quasi-gold standard based on records that contained national ID numbers. We calculated the probabilities that the algorithm could correctly identify a true match (sensitivity) and the probabilities that the match identified by the algorithm was truly a match (positive predictive value) using the two approaches. ResultsBetween 2015 and 2020, 55,077 patients were recorded in the HIV treatment database and 386,577 laboratory tests were recorded in the laboratory service database. The sensitivity and positive predictive value in the manually reviewed data were estimated to be 94.4% and 81.2% respectively. The sensitivity and positive predictive value in the matched individual identifiers were estimated to be 99.6% and 99.7% respectively. ConclusionsRecords can be linked successfully, but estimated performance of record linkage depends on the validation set used. Manually reviewed data contain noise and may underestimate performance, while national ID numbers may overestimate performance due to non-random patterns of missingness.

Enabling Fast and Accurate Record Linkage of Large-Scale Health-Related Administrative Databases Through a DNA-Encoded Approach.

ObjectivePublic health research frequently requires the integration of information from different data sources. However, errors in the records and the high computational costs involved make linking large administrative databases using record linkage (RL) methodologies a major challenge. We present Tucuxi-BLAST, a versatile tool for probabilistic RL that utilizes a DNA-encoded approach to encrypt, analyze and link massive administrative databases. Materials and MethodsTucuxi-BLAST encodes the identification records into DNA. BLASTn algorithm is then used to align the sequences between databases. We tested and benchmarked on a simulated database containing records for 300 million individuals and also on four large administrative databases containing real data on Brazilian patients. ResultsOur method was able to overcome misspellings and typographical errors in administrative databases. In processing the RL of the largest simulated dataset (200k records), the state-of-the art method took 5 days and 7 hours to perform the RL, while Tucuxi-BLAST only took 23 hours. When compared with five existing RL tools applied to a gold-standard dataset from real health-related databases, Tucuxi-BLAST had the highest accuracy and speed. DiscussionBy repurposing genomic tools, researchers are able to perform subject tracing across multiple large epidemiological databases using a regular laptop. ConclusionTucuxi-BLAST can improve data-driven medical research and provide a fast and accurate way to link individual information across several administrative databases.

All-cause and cause-specific mortality among people with severe mental illness in Brazil's public health system, 2000–15: a retrospective study

SummaryBackgroundPeople with severe mental illness have a mortality rate higher than the general population, living an average of 10–20 years less. Most studies of mortality among people with severe mental illness have occurred in high-income countries (HICs). We aimed to estimate all-cause and cause-specific relative risk (RR) and excess mortality rate (EMR) in a nationwide cohort of inpatients with severe mental illness compared with inpatients without severe mental illness in a middle income country, Brazil.MethodsThis national retrospective cohort study included all patients hospitalised through the Brazilian Public Health System (Sistema Único de Saúde [SUS]-Brazil) between Jan 1, 2000, and April 21, 2015. Probabilistic and deterministic record linkages integrated data from the Hospital Information System (Sistema de informações Hospitalares) and the National Mortality System (Sistema de Informação sobre Mortalidade). Follow-up duration was measured from the date of the patients’ first hospitalisation until their death, or until April 21, 2015. Severe mental illness was defined as schizophrenia, bipolar disorder, or depressive disorder by ICD-10 codes used for the admission. RR and EMR were calculated with 95% CIs, comparing mortality among patients with severe mental illness with those with other diagnoses for patients aged 15 years and older. We redistributed deaths using the Global Burden of Diseases, Injuries, and Risk Factors Study methodology if ill-defined causes of death were stated as an underlying cause.FindingsFrom Jan 1, 2000, to April 21, 2015, 72 021 918 patients (31 510 035 [43·8%] recorded as male and 40 974 426 [56·9%] recorded as female; mean age 41·1 (SD 23·8) years) were admitted to hospital, with 749 720 patients (372 458 [49·7%] recorded as male and 378 670 [50·5%] as female) with severe mental illness. 5 102 055 patient deaths (2 862 383 [56·1%] recorded as male and 2 314 781 [45·4%] as female) and 67 485 deaths in patients with severe mental illness (39 099 [57·9%] recorded as male and 28 534 [42·3%] as female) were registered. The RR for all-cause mortality in patients with severe mental illness was 1·27 (95% CI 1·27–1·28) and the EMR was 2·52 (2·44–2·61) compared with non-psychiatric inpatients during the follow-up period. The all-cause RR was higher for females and for younger age groups; however, EMR was higher in those aged 30–59 years. The RR and EMR varied across the leading causes of death, sex, and age groups. We identified injuries (suicide, interpersonal violence, and road injuries) and cardiovascular disease (ischaemic heart disease) as having the highest EMR among those with severe mental illness. Data on ethnicity were not available.InterpretationIn contrast to studies from HICs, inpatients with severe mental illness in Brazil had high RR for idiopathic epilepsy, tuberculosis, HIV, and acute hepatitis, and no significant difference in mortality from cancer compared with inpatients without severe mental illness. These identified causes should be addressed as a priority to maximise mortality prevention among people with severe mental illness, especially in a middle-income country like Brazil that has low investment in mental health.FundingBill and Melinda Gates Foundation, Fundação de Amparo a Pesquisa do Estado de Minas Gerais, FAPEMIG, and the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brasil.

Tucuxi-BLAST: Enabling fast and accurate record linkage of large-scale health-related administrative databases through a DNA-encoded approach.

BackgroundPublic health research frequently requires the integration of information from different data sources. However, errors in the records and the high computational costs involved make linking large administrative databases using record linkage (RL) methodologies a major challenge.MethodsWe present Tucuxi-BLAST, a versatile tool for probabilistic RL that utilizes a DNA-encoded approach to encrypt, analyze and link massive administrative databases. Tucuxi-BLAST encodes the identification records into DNA. BLASTn algorithm is then used to align the sequences between databases. We tested and benchmarked on a simulated database containing records for 300 million individuals and also on four large administrative databases containing real data on Brazilian patients.ResultsOur method was able to overcome misspellings and typographical errors in administrative databases. In processing the RL of the largest simulated dataset (200k records), the state-of-the-art method took 5 days and 7 h to perform the RL, while Tucuxi-BLAST only took 23 h. When compared with five existing RL tools applied to a gold-standard dataset from real health-related databases, Tucuxi-BLAST had the highest accuracy and speed. By repurposing genomic tools, Tucuxi-BLAST can improve data-driven medical research and provide a fast and accurate way to link individual information across several administrative databases.

Administrative databases as a tool for identifying healthcare demand and costs in an over-one million population

&#x0D; &#x0D; Background: the aim of this study was to assess healthcare demand of specific groups of population and their costs borne by Italian Health System, using healthcare administrative databases.&#x0D; Methods: demographic, clinical and economic data were obtained from datasets available at the Regional Health System, combined into a data warehouse (DENALI), using a probabilistic record linkage to optimize the data matching process. The study population consisted of more than 1 million people registered in 2005 at one Local Healthcare Unit of Lombardy. Eight different segments were identified. Costs occurring in 2005 for hospital admissions, drug prescriptions, outpatient medical specialist visits were quantified in each segment.&#x0D; Results: healthy people accounted for 53% of the population and cost € 180 per-capita. Subjects with only one chronic disease made up 16% of the population and cost € 916 per-capita, those affected by several chronic diseases accounted for 13% and cost € 3 457 per-capita. Hospitalizations were the cost driver in five segments, ranging from 42% to 89% of total expenditures. Outpatient visits were the cost driver among healthy subjects (54%) and those with a possible chronic disease (42%), while drug costs ranged between 4% (“acute event”) and 32% (“one chronic disease”). Overall, healthcare cost was € 809 per-capita.&#x0D; Conclusions: healthcare costs were mainly determined by people affected by chronic conditions, even if “healthy people” ranked third for total expenditure. These costs need an appropriate identification of healthcare demand, that could be efficiently monitored through the use of administrative databases.&#x0D; &#x0D;

Evaluation of string comparators for record linkage in Chinese environment

In Chinese record linkage, the similarity measurement of short text fields such as identifying information has an important influence on the effect of linkage. In this paper, we compared the performance of six different string comparators for Chinese record linkage, including Hamming, Levenshtein, Jaro, Overlap Coefficient, Cosine and Ratcliff–Obershelp similarity. At the same time, the expectation-maximization algorithm was used to perform the probabilistic record linkage. The results showed that compared with other string comparators, Jaro similarity not only gets higher precision and recall, but also has the least computational time. All in all, this comparator can be used if we focus on string spelling and pronunciation without considering semantic similarity.

Using probabilistic record linkage and propensity-score matching to identify a community-based comparison population.

In retrospective cohort studies of interventions disseminated to communities, it is challenging to find comparison groups with high-quality data for evaluation. We present one methodological approach as part of our study of birth outcomes of second-born children in a home visiting (HV) program targeting first-time mothers. We used probabilistic record linkage to link Connecticut's Nurturing Families Network (NFN) HV program and birth-certificate data for children born from 2005 to 2015. We identified two potential comparison groups: a propensity-score-matched group from the remaining birth certificate sample and eligible-but-unenrolled families. An analysis of interpregnancy interval (IPI) is presented to exemplify the approach. We identified the birth certificates of 4822 NFN families. The propensity-score-matched group had 14,219 families (3-to-1 matching) and we identified 1101 eligible-but-unenrolled families. Covariates were well balanced for the propensity-score-matched group, but poorly balanced for the eligible-but-unenrolled group. No program effect on IPI was found. By combining propensity-score matching and probabilistic record linkage, we were able to retrospectively identify relatively large comparison groups for quasi-experimental research. Using birth certificate data, we accessed outcomes for all of these individuals from a single data source. Multiple comparison groups allow us to confirm findings when each method has some limitations. Other researchers seeking community-based comparison groups could consider a similar approach.

Cohort profile: the South African HIV Cancer Match (SAM) Study, a national population-based cohort

PurposeThe South African HIV Cancer Match (SAM) Study is a national cohort of people living with HIV (PLWH). It was created using probabilistic record linkages of routine laboratory records of...