A Scoping Review of Data Sources for the Conduct of Policy-Relevant Substance Use Research

Objective Existing administrative and survey data are critical for understanding the effects of exigent policies on population health outcomes related to opioid, cannabis, and other substance use disorders (SUDs). The objective of this study was to determine the state of the data available for evaluating SUD-related health outcomes. Methods We performed a scoping review of national and state government data sources to measure and evaluate the effects of state policy changes on substance use and SUD-related health outcomes and health care use. We used Massachusetts as a case study for availability of relevant state-level data as well as national datasets with state-level indicators available to measure outcomes. We compared key features of each dataset to assess their usefulness for research and policy evaluation. We conducted our review during November 2018–March 2019, and we updated data availability as of March 2019 for all data sources. Results We identified 11 survey datasets, 12 national administrative datasets, and 10 state administrative datasets as being suitable for policy-relevant research and practice purposes. These datasets varied substantially in their usefulness for evaluation and research. Despite substantial data limitations, including prohibitive regulatory and monetary costs to obtain the data and limited availability, these data can be mined to examine a diversity of policy-relevant questions. Conclusions Findings provide a comprehensive resource for using survey and administrative data to evaluate the health effects of SUD-related policies and interventions. The construction of state-level public health data warehouses or record linkage projects connecting individual-level information in state data sources is valuable for analyzing the effects of policy changes. Understanding strengths and limitations of available data sources is important for ongoing research and evaluation.

A bespoke data linkage of an IVF clinical quality registry to population health datasets; methods and performance

IntroductionAssisted reproductive technologies (ART), such as in-vitro fertilisation (IVF), have revolutionised the treatment of infertility, with an estimated 8 million babies born worldwide. However, the long-term health outcomes for women and their offspring remain an area of concern. Linking IVF treatment data to long-term health data is the most efficient method for assessing such outcomes. ObjectivesTo describe the creation and performance of a bespoke population-based data linkage of an ART clinical quality registry to state-based and national administrative datasets. MethodsThe linked dataset was created by deterministically and probabilistically linking the Australia and New Zealand Assisted Reproduction Database (ANZARD) to New South Wales (NSW) and Australian Capital Territory (ACT) administrative datasets (performed by NSW Centre for Health Record Linkage (CHeReL)) and to national claims datasets (performed by Australian Institute of Health and Welfare (AIHW)). The CHeReL's Master Linkage Key (MLK) was used as a bridge between ANZARD's partially identifiable patient data (statistical linkage key) and NSW and ACT administrative datasets. CHeReL then provided personal identifiers to the AIHW to obtain national content data. The results of the linkage were reported, and concordance between births recorded in ANZARD and perinatal data collections (PDCs) was evaluated. ResultsOf the 62,833 women who had ART treatment in NSW or ACT, 60,419 could be linked to the CHeReL MLK (linkage rate: 96.2%). A reconciliation of ANZARD-recorded births among NSW residents found that 94.2% (95% CI: 93.9--94.4%) of births were also recorded in state/territory-based PDCs. A high concordance was found in plurality status and birth outcome (≥99% agreement rate, Cohen's kappa ranged: 0.78--0.98) between ANZARD and PDCs. ConclusionThe data linkage resource demonstrates that high linkage rates can be achieved with partially identifiable data and that a population spine, such as the CHeReL's MLK, can be successfully used as a bridge between clinical registries and administrative datasets.

918Building the evidence base for australia’s veteran population

Background There is a wealth of existing research on Australia’s veteran population. However, much of the existing research seeks to answer questions on specific issues, veterans of specific conflicts, or of veterans who use the Department of Veterans’ Affairs services, with little to no information beyond these specific populations. Methods Several large administrative datasets including the Pharmaceutical Benefits Schedule (PBS), Causes of Death and Specialist Homelessness Services were linked with Defence information to enable analysis of veterans’ information against a range of topics, such as cause of death, health service and homelessness service use. Comparisons were made between the Australian Population as a whole and the veteran population. Results Results show some similarities and some differences between the veteran population and the whole Australian population. Results are disaggregated between those who are currently serving, in the reserves and ex-serving as well as by service type and other variables. Conclusions These results have helped to improve what is known about the broader veteran community. They provide important context for policy makers to understand how best to service the veteran community that was previously unknown, and help identify areas for more detailed research and exploration. Key messages Data linkage provides an important opportunity to understand the specific health and welfare needs of veterans who are otherwise difficult to identify in large administrative datasets. Comparison of veterans’ health and welfare service use with the broader Australian population provides important information for policy makers.

Predicting postoperative surgical site infection with administrative data: a random forests algorithm

Background Since primary data collection can be time-consuming and expensive, surgical site infections (SSIs) could ideally be monitored using routinely collected administrative data. We derived and internally validated efficient algorithms to identify SSIs within 30 days after surgery with health administrative data, using Machine Learning algorithms. Methods All patients enrolled in the National Surgical Quality Improvement Program from the Ottawa Hospital were linked to administrative datasets in Ontario, Canada. Machine Learning approaches, including a Random Forests algorithm and the high-performance logistic regression, were used to derive parsimonious models to predict SSI status. Finally, a risk score methodology was used to transform the final models into the risk score system. The SSI risk models were validated in the validation datasets. Results Of 14,351 patients, 795 (5.5%) had an SSI. First, separate predictive models were built for three distinct administrative datasets. The final model, including hospitalization diagnostic, physician diagnostic and procedure codes, demonstrated excellent discrimination (C statistics, 0.91, 95% CI, 0.90–0.92) and calibration (Hosmer-Lemeshow χ2 statistics, 4.531, p = 0.402). Conclusion We demonstrated that health administrative data can be effectively used to identify SSIs. Machine learning algorithms have shown a high degree of accuracy in predicting postoperative SSIs and can integrate and utilize a large amount of administrative data. External validation of this model is required before it can be routinely used to identify SSIs.

Sociodemographic associations of geographic variation in paediatric tonsillectomy and adenoidectomy

Geographic variation of paediatric tonsillectomy, with or without adenoidectomy, (A/T) has been described since the 1930s until today but no studies have investigated the factors associated with this variation. This study described the geographical distribution of paediatric A/T across the state of Victoria, Australia, and investigated area-level factors associated with this variation. We used linked administrative datasets capturing all paediatric A/T performed between 2010 and 2015 in Victoria. Surgery data were collapsed by patient residence to the level of Local Government Area. Regression models were used to investigate the association between likelihood of surgery and area-level factors. We found a 10.2-fold difference in A/T rates across the state, with areas of higher rates more in regional than metropolitan areas. Area-level factors associated with geographic variation of A/T were percentage of children aged 5–9 years (IRR 1.07, 95%CI 1.01–1.14, P = 0.03) and low English language proficiency (IRR 0.95, 95% CI 0.90–0.99, P = 0.03). In a sub-population analysis of surgeries in the public sector, these factors were low maternal educational attainment (IRR 1.09, 95% CI 1.02–1.16, P < 0.001) and surgical waiting time (IRR 0.99635 95% CI 0.99273–0.99997, P = 0.048). Identifying areas of focus for improvement and factors associated with geographic variation will assist in improving equitable provision of paediatric A/T and decrease variability within regions.

Best Paper Selection

Powell KR, Deroche CB, Alexander GL. Health Data Sharing in US Nursing Homes: A Mixed Methods Study. https://www.jamda.com/article/S1525-8610(20)30197-3/fulltext Cappetta K, Lago L, Potter J, Phillipson L. Under-coding of dementia and other conditions indicates scope for improved patient management: A longitudinal retrospective study of dementia patients in Australia. https://journals.sagepub.com/doi/abs/10.1177/1833358319897928?journalCode=himd Sheriffdeen A, Millar JL, Martin C, Evans M, Tikellis G, Evans SM. (Dis)concordance of comorbidity data and cancer status across administrative datasets, medical charts, and self-reports. https://bmchealthservres.biomedcentral.com/articles/10.1186/s12913-020-05713-5

A Note on Posttreatment Selection in Studying Racial Discrimination in Policing

We discuss some causal estimands that are used to study racial discrimination in policing. A central challenge is that not all police–civilian encounters are recorded in administrative datasets and available to researchers. One possible solution is to consider the average causal effect of race conditional on the civilian already being detained by the police. We find that such an estimand can be quite different from the more familiar ones in causal inference and needs to be interpreted with caution. We propose using an estimand that is new for this context—the causal risk ratio, which has more transparent interpretation and requires weaker identification assumptions. We demonstrate this through a reanalysis of the NYPD Stop-and-Frisk dataset. Our reanalysis shows that the naive estimator that ignores the posttreatment selection in administrative records may severely underestimate the disparity in police violence between minorities and whites in these and similar data.

Identifying pre-existing dementia in older adults diagnosed with cancer using a national claims database.

e18678 Background: Older adults over the age of 65 represent the majority of patients diagnosed with (60%), among them, 15-30% have a pre-existing Alzheimer’s disease or related dementia (ADRD) that puts them at higher risk for over and under treatment. Studying the role of pre-existing ADRD in cancer patients is vital to understanding treatment planning behavior, patterns of health care utilization, and adverse treatment outcomes. Massive administrative datasets, or “big data” represent the information rich environment that is useful for this endeavor. Methods: Our study utilized a clinically validated algorithm to assess the prevalence of pre-existing ADRD and cancer across six cancer types. We utilized the SEER-Medicare dataset for analyzing the study years 2004-2015 (N = 337 932). We extracted ICD-9 codes to identify ADRD using the Centers for Medicaid Services Chronic Conditions Warehouse (CCW) algorithm. In sensitivity analysis we compared the prevalence of ADRD+Cancer using the NCI (2014) and CCW algorithms. Results: We found a significant difference between the two algorithms (p < .0001) and a higher overall prevalence of comorbid ADRD+Cancer using the CCW (6.6%). Additionally, we found ADRD+Cancer prevalence was significantly higher among racial and ethnic subgroups compared to White and unstaged tumors compared with any numbered American Joint Committee on Cancer (AJCC) stages (p < .0001). Conclusions: Using a clinically validated algorithm we were able to identify more cases of ADRD+Cancer in big data. This figure remains underestimated for ADRD+cancer compared to clinically-validated studies. Further research into the validation approach and codes that are used for ADRD classification can improve how we identify ADRD in massive administrative data. This is critical given the growing population of diverse older adults in the U.S.