Probabilistic Linkage Methodology Research Articles

Linkage of public health and all payer claims data for population‐level opioid research

Our objective is to describe how we combine, at an individual level, multiple administrative datasets to create a Comprehensive Opioid Risk Registry (CORR). The CORR will characterize the role that individual characteristics, household characteristics, and community characteristics have on an individual's risk of opioid use disorder or opioid overdose. Study data sources include the voluntary Oregon All Payer Claims Database (APCD), American Community Survey Census Data, Oregon Death Certificate data, Oregon Hospital Discharge Data (HDD), and Oregon Prescription Drug Monitoring (PDMP) Data in 2013-2018. To create the CORR we first prepared the APCD data set by cleaning and geocoding addresses, creating a community grouper and adding census indices, creating household grouper, and imputing patient race. Then we deployed a probabilistic linkage methodology to incorporate other data sources maintaining compliance with strict data governance regulations. Administrative datasets were obtained through an executed data use agreement with each data owner. The APCD served as the population universe to which all other data sources were linked. There were 3 628 992 unique people in the APCD over the entire study period. We identified 968 767 unique households in 2013 and 1 209 236 in 2018, and geocoded patient addresses representing all census tracts in Oregon. Census, death certificate, HDD, and PDMP datasets were successfully linked to this population universe. This methodology can be replicated in other states and may also apply to a broad array of health services research topics.

Probabilistic Linkage of Randomized Controlled Trial Data to Administrative Claims: A Case Study of Patients from Baricitinib Clinical Trials.

IntroductionThe aim of this work is to assess the feasibility of probabilistically linking randomized controlled trial (RCT) data to claims data in a real-world setting to inform future rheumatoid arthritis (RA) research.MethodsThis retrospective cohort study utilized IQVIA’s Patient Centric Medical Claims (Dx) Database, IQVIA’s Longitudinal Prescription Claims (LRx) Database, and Lilly’s baricitinib RCT data from a sample of patients that consented to the linkage of their de-identified insurance claims to their de-identified RCT data. Patients were initially matched on age, gender, and three-digit ZIP code of the provider and further matched according to a point scoring system using additional clinical variables.ResultsA total of 245 patients from 49 US clinical trial sites were eligible for the study and 78 (31.8%) of these patients consented to participate. Of the 78 consented patients, 69 (88%) were successfully matched on age, gender, and three-digit ZIP code of the provider. Of the 69 patients successfully matched on age, gender, and three-digit ZIP code of the provider, 44 (63.8%) had at least one sufficient match using the point scoring system. Of these 44, 23 (52.3%) patients matched at a ratio of one RCT patient to one Dx/LRx patient, 11 (25.0%) at a ratio of 1:2, 7 (15.9%) at a ratio of 1:3 and three (6.8%) at a ratio of 1:4 or greater. To further improve match ratios, a variable hierarchy was applied to the 18 RCT patients with 2–3 matches. Overall, 38 of the 78 (48.7%) consented RCT patients were successfully matched 1:1 to claims database patients.ConclusionsThis probabilistic linkage methodology demonstrates the feasibility, at a moderate linkage rate, of linking patients from RCTs to real-world data, which can provide a means to assess additional information not usually collected within or following a clinical trial.

Comparison of State Hospital Visits With Housing and Urban Development Estimates of Homeless: Illinois, 2011-2018.

Objectives. To assess the value of hospital records in augmenting information on homelessness counts at a state level.Methods. We used data from the Illinois Hospital Discharge Database (2011-2018) to identify outpatients and inpatients identified as affected by homelessness. We used probabilistic linkage methodology to estimate unique individuals rather than visits and compared them with US Department of Housing and Urban Development annual estimates of homelessness based on point-in-time counts.Results. Department of Housing and Urban Development point-in-time estimates indicate a substantial decline of approximately 24% in homelessness in Illinois; however, estimates of unique individuals visiting the hospital with a code for homelessness more than doubled in this same period.Conclusions. Other data sources, such as hospital records, are increasingly able to identify and report information related to homelessness. Using these additional data sources may help to augment the Department of Housing and Urban Development point-in-time estimates to provide more accurate estimates of homelessness that are used to direct resources and assess policy and support services for those affected by homelessness.

Challenges of Using Probabilistic Linkage Methodology to Characterize Post-Cardiac Arrest Care in Michigan

Background: To improve survival of patients resuscitated from out of hospital cardiac arrest (OCHA), data is needed to assess and improve inpatient post-resuscitation care. Our objective was to apply probabilistic linkage methodology to link EMS and inpatient databases and evaluate whether it may be used to describe post-arrest care in Michigan. Methods: We performed a retrospective study to describe post-cardiac arrest care in adult OHCA patients who were transported to Michigan hospitals from July 1, 2010, to June 30, 2013. Using probabilistic linkage methodology we linked two databases, the Michigan EMS Information System (MI_EMSIS) and the Michigan Inpatient Database (MIDB), which describes inpatient care and outcome of all admissions. Rates of case incidence and survival were compared to published literature. We compared the linked dataset to existing cardiac arrest databases from three counties to evaluate the quality of this linkage. Results: Multiple iterations of match strategies were used to create a linked EMS-inpatient dataset. There were 12,838 MI_EMSIS cardiac arrest records of which 1,977 were matched with MIDB records, identifying them as surviving to hospital admission. Of these 590 (30.0%) survived to hospital discharge. The annual survival incidence/100,000 population to admission was 6.93/100,000 and survival incidence to discharge was 2.1/100,000. The matched dataset was compared to county databases identified a limited sensitivity [48.2%, 95% CI 42.1%–55.3%)] and positive predictive value [64.4%, 95% CI 56.8%–71.3%)]. Conclusion: Use of the MI_EMSISEMS database and the Michigan Inpatient database was feasible and produced rates of cardiac arrest admission and survival rates similar to published literature. This process yielded a limited match compared to existing county cardiac arrest databases. We conclude that such a linked dataset is useful for descriptive purposes but not as a population based dataset to evaluate statewide post-cardiac arrest care.

States Monitoring Assisted Reproductive Technology (SMART) Collaborative: Data Collection, Linkage, Dissemination, and Use

Assisted reproductive technology (ART) refers to fertility treatments in which both eggs and sperm are handled outside the body. The Centers for Disease Control and Prevention (CDC) oversees the National ART Surveillance System (NASS), which collects data on all ART procedures performed in the United States. The NASS, while a comprehensive source of data on ART patient demographics and clinical procedures, includes limited information on outcomes related to women's and children's health. To examine ART-related health outcomes, CDC and three states (Massachusetts, Florida, and Michigan) established the States Monitoring ART (SMART) Collaborative to evaluate maternal and perinatal outcomes of ART and improve state-based ART surveillance. To date, NASS data have been linked with states' vital records, disease registries, and hospital discharge data with a linkage rate of 90.2%. The probabilistic linkage methodology used in the SMART Collaborative has been validated and found to be both accurate and efficient. A wide breadth of applied research within the Collaborative is planned or ongoing, including examinations of the impact of insurance mandates on ART use as well as the relationships between ART and birth defects and cancer, among others. The SMART Collaborative is working to improve state-based ART surveillance by developing state surveillance plans, establishing partnerships, and conducting data analyses. The SMART Collaborative has been instrumental in creating linked datasets and strengthening epidemiologic and research capacity for improving maternal and infant health programs and evaluating the public health impact of ART.

Abstract TP245: Impact Of A County-wide Prehospital Destination Protocol On Thrombolytic Rates For Acute Ischemic Stroke (AIS)

Background: Higher IV t-PA rates were observed statewide after regionalization of stroke. Objectives: We describe the IV t-PA rates in ambulance transported acute ischemic stroke (AIS) patients before and during implementation of a county wide prehospital stroke center destination protocol(SDP)and examine the association between SDP and IV t-PA rates in ambulance transported AIS patients. Methods: This is a cross-sectional observational study of patients with a hospital based diagnosis of AIS identified using validated ICD- 9 codes. Patient records from 2005-2007 were obtained from the discharge abstract file of the statewide administrative database and were linked to the prehospital electronic records using patient level identifiers and probabilistic linkage methodology. Thrombolytic use for AIS was identified using the procedure codes in the discharge database. We excluded direct admissions and inter-facility transfers. IV t_PA rate by year was calculated.Logistic regression was used to determine association between SDP and the IV t_PA rate in AIS patients after controlling for patient and hospital demographics, stroke center designation and teaching status of the hospital, patient residence and day of the week using logistic regression. Data analysis was performed using SAS 9.2 Results: During a 3 year period, 6181 patients with a primary or secondary diagnosis of stroke were transported by ambulance. Mean age at time of admission was 74 (+/-15) years; 54% (n=3312) were females and 63% (n=3870) were whites. Majority of patients were treated at stroke centers 4132 (70%) and 6005 (97%) were treated at community hospitals. Among ambulance transports, IV t-PA rate did not increase during the implementation of the stroke center designation protocol (pre protocol phase 2.82%, post protocol phase 2.85%, p value 0.95). After controlling for patient demographics, stroke center status, teaching status of the hospital and the weekend effect, SDP implementation was not independently associated with increased rate of IV t-PA in AIS (OR 0.96, 95 CI 0.63-1.47). Conclusions: Among ambulance transported patients, our preliminary findings do not show an increase in thrombolytic rates during implementation of a stroke center destination program.

Abstract 154: Acute Stroke Recognition by Paramedics after Regionalization of Stroke Care: Outcomes Based Study

Background: Early evidence supports preferential transport of patients with stroke symptoms to primary stroke centers. While validated stroke tools exist for screening of stroke symptoms in the prehospital setting, system wide triage performance of prehospital providers in a regionalized system has not been reported. The objective of this study is to assess the diagnostic ability of prehospital providers, before and after regionalization of care, using outcomes based approach. Methods: This is a cross-sectional study of all patients who were transported to hospitals in two Northern California counties by providers of a single EMS agency during a three year period. One county remained non-regionalized (NR) during the study period and the other initiated and completed regionalization(R) of the system during the study period. Patient demographic data, prehospital provider clinical assessment was obtained from the computerized prehospital transport records and physician diagnosis was obtained from statewide administrative patient discharge data. The data sources were linked using probabilistic linkage methodology. Patients ≥18 years of age with validated ICD- 9 code for stroke were included. We excluded inter-facility transports and direct admissions. Sensitivity, specificity and predictive values for were determined before and after implementation of regionalization. Data analysis was performed using SAS version 9.2. Results: The total number of medical related EMS transports for 3 years was 310,731 and the number of patient discharges with a primary diagnosis of stroke was 10,298. We were able to link 3736 stroke records which indicate EMS use by 36% (3736/10,298) stroke patients. The sensitivity, specificity, PPV and NPV in the pre-regionalization phase was 28%, 80%, 53%, 58% and during the implementation phase of regionalization was 39% 78%,57% and 63% (p <0.05). The performance in the NR County during the entire period was 23%, 76%, 40% and 58%. Conclusions: Diagnostic accuracy remained low although improved prehospital provider performance was observed after regionalization of stroke care.

Abstract 164: Treatment Differences in Acute Stroke By Mode of Transport

Background: EMS (Emergency Medical Services) is an important component of the stroke chain of survival. While in hospital stroke time targets have shown to be better when prehospital providers provide advance notification, population-wide studies on thrombolysis rates (IV t-PA) for those who arrive by EMS is sparse. Objective: To examine differences in treatment rates for acute stroke by mode of transport. Methods: This is a cross-sectional study of all patients who were transported to hospitals in two Northern California counties by providers of a single EMS agency during a three year period. Patient demographic data, prehospital provider clinical assessment was obtained from the computerized prehospital transport records and patient location, hospital demographics, physician diagnosis and treatment rates were obtained from statewide administrative patient discharge data. The data sources were linked using probabilistic linkage methodology. Patients ≥18 years of age with validated ICD- 9 code for stroke were included. We excluded inter-facility transports and direct admissions. Results: Of 10,456 patients who had a hospital based discharge diagnosis of stroke, 3787 (36%) were transported by EMS. Mean age at the time of admission was 75 years (+/- 14); 55% (2093) were females, 65% (2471) were whites and 86% (3247) were Hispanics. Most of the patients had Medicare (72%, 2737) and 92% (3471) were transported from home. Majority of the patients were treated at stroke centers [n=3014, (80%)]) and at community hospitals [n=3664, (97%)]. Of 3757 patients with a primary diagnosis of stroke, 4% (150) were treated with IV t-PA. After controlling for covariates, patients transported by EMS had higher odds of treatment with IV t-PA (OR 2.6, 95 CI 1.9-3.3). Treatment at stroke centers (OR 1.5, 95 CI 1-2.2) and academic centers were independently associated with treatment rates (OR 2.4, 95 CI 1.6-3.6). Conclusions: Arrival by EMS to emergency department is associated with higher treatment rates with thrombolytics for acute stroke patients and efforts should be targeted to improve use of EMS for stroke.

Probabilistic linkage in household survey on hospital care usage.

To evaluate the potential advantages and limitations of the use of the Brazilian hospital admission authorization forms database and the probabilistic record linkage methodology for the validation of reported utilization of hospital care services in household surveys. A total of 2,288 households interviews were conducted in the county of Duque de Caxias, Brazil. Information on the occurrence of at least one hospital admission in the year preceding the interview was obtained from a total of 10,733 household members. The 130 records of household members who reported at least one hospital admission in a public hospital were linked to a hospital database with 801,587 records, using an automatic probabilistic approach combined with an extensive clerical review. Seventy-four (57%) of the 130 household members were identified in the hospital database. Yet only 60 subjects (46%) showed a record of hospitalization in the hospital database in the study period. Hospital admissions due to a surgery procedure were significantly more likely to have been identified in the hospital database. The low level of concordance seen in the study can be explained by the following factors: errors in the linkage process; a telescoping effect; and an incomplete record in the hospital database. The use of hospital administrative databases and probabilistic linkage methodology may represent a methodological alternative for the validation of reported utilization of health care services, but some strategies should be employed in order to minimize the problems related to the use of this methodology in non-ideal conditions. Ideally, a single identifier, such as a personal health insurance number, and the universal coverage of the database would be desirable.

Visual impairment and driving restrictions. Wick JP, Vernon DD. Dig J Ophthalmol 2002;8:1. (http://www.djo.harvard.edu/meei/OA/Wick/index.html)

The authors sought to determine the adverse incident rates of visually impaired drivers at specific levels of driving restriction and to compare these rates to unimpaired, unrestricted drivers. Visually impaired drivers were identified from statewide databases containing over 1.7 million individuals. Probabilistic linkage methodology was used to link databases containing information on drivers’ licenses, medical conditions, citations and crashes. Exposed drivers with visual impairment (n = 1,527) were compared to an unrestricted, unimpaired cohort (n = 2,939) at three levels of driving restriction and visual impairment (low, moderate, high) for rates of citations, crashes, and at-fault crashes. Exposed drivers were case matched to unexposed cohorts based on age, gender, and geographic location, and the degree of driving restrictions included limitations on speed, area, or time of day. The outcomes were expressed as odds ratio for each of three levels of restriction. When compared to the unimpaired, unrestricted drivers, visually impaired drivers demonstrated a significantly increased number of citations for two of the three levels of driving restriction; significantly increased number of crashes for two of the three levels of driving restriction; and a significantly increased number of at-fault crashes for two of the three levels of driving restriction. The highest odds ratios were associated with the most impaired and most restricted drivers. The authors conclude that visually impaired, restricted drivers are still involved in more motor vehicle crashes and citations than unimpaired, unrestricted drivers, but this increase is modest for drivers with moderate visual impairment and restriction. However, the higher rate of adverse incidents in the most impaired subjects with the most restrictions warrants further investigation.—Thomas J. Liesegang

Evaluating the crash and citation rates of Utah drivers licensed with medical conditions, 1992–1996

Background: Medical problems may affect the ability to drive motor vehicles, and programs that control the issuing of driver licenses to individuals with medical conditions exist in most states. The main activity of these programs is the imposition of restrictions upon the driving privileges of individuals with medical conditions that are deemed to pose some risk to public safety. However, little is known about the effectiveness of these licensing programs. Objective: The objective of this study was to compare the rates of adverse driving events (crash, at-fault crash and citations) experienced by drivers licensed with medical conditions to those of age-, sex- and location-matched controls. Separate comparisons were made for drivers reporting medical conditions licensed with full driving privileges, and those with restricted driving privileges (e.g. speed, area and time of day). Design: Retrospective case-control. Methods: The study population was all drivers licensed in the state of Utah who reported a medical condition on their driver license application, over the 5-year period 1992–1996. Drivers enter the program by self-reporting their medical problems. Control drivers were chosen from the entire population of drivers licensed in Utah for the same period. Information on driver license status, participation in the medical conditions program, citations, involvement in crashes, and death certificate data was obtained from the relevant state agencies. Probabilistic linkage methodology was used to link the records in these disparate databases for eventual analysis. Rates of citation, crashes and at-fault crashes, expressed as events per 10 000 license days, were calculated separately for program drivers and their corresponding control groups for each medical condition category and restriction status. These data were used to determine an estimate of relative risk (RR) and 95% confidence intervals. Results: As a group, medical conditions drivers had modestly elevated rates of adverse driving events compared with control drivers (RR 1.09–1.74). Rates in the largest medical category, ‘cardiovascular conditions’, were not higher than controls. Rates were higher than control for some conditions, such as ‘alcohol’ and ‘learning and memory’, for some adverse events (RR 2.2–5.75). Drivers with more than one medical condition appeared comparable to the general group of medical conditions program drivers. Conclusions: Drivers in Utah medical conditions program had modestly elevated rates of adverse driving events compared to matched controls. Possible underreporting of medical conditions and accurate assessment of exposure rates are potential weaknesses in the program.