Integrating human services and criminal justice data with claims data to predict risk of opioid overdose among Medicaid beneficiaries: A machine-learning approach.

Health system data incompletely capture the social risk factors for drug overdose. This study aimed to improve the accuracy of a machine-learning algorithm to predict opioid overdose risk by integrating human services and criminal justice data with health claims data to capture the social determinants of overdose risk. This prognostic study included Medicaid beneficiaries (n = 237,259) in Allegheny County, Pennsylvania enrolled between 2015 and 2018, randomly divided into training, testing, and validation samples. We measured 290 potential predictors (239 derived from Medicaid claims data) in 30-day periods, beginning with the first observed Medicaid enrollment date during the study period. Using a gradient boosting machine, we predicted a composite outcome (i.e., fatal or nonfatal opioid overdose constructed using medical examiner and claims data) in the subsequent month. We compared prediction performance between a Medicaid claims only model to one integrating human services and criminal justice data with Medicaid claims (i.e., integrated model) using several metrics (e.g., C-statistic, number needed to evaluate [NNE] to identify one overdose). Beneficiaries were stratified into risk-score decile subgroups. The samples (training = 79,087, testing = 79,086, validation = 79,086) had similar characteristics (age = 38±18 years, female = 56%, white = 48%, having at least one overdose = 1.7% during study period). Using the validation sample, the integrated model slightly improved on the Medicaid claims only model (C-statistic = 0.885; 95%CI = 0.877-0.892 vs. C-statistic = 0.871; 95%CI = 0.863-0.878), with small corresponding improvements in the NNE and positive predictive value. Nine of the top 30 most important predictors in the integrated model were human services and criminal justice variables. Using the integrated model, approximately 70% of individuals with overdoses were members of the top risk decile (overdose rates in the subsequent month = 47/10,000 beneficiaries). Few individuals in the bottom 9 deciles had overdose episodes (0-12/10,000). Machine-learning algorithms integrating claims and social service and criminal justice data modestly improved opioid overdose prediction among Medicaid beneficiaries for a large U.S. county heavily affected by the opioid crisis.

The Effect of Incomplete Death Certificates on Estimates of Unintentional Opioid-Related Overdose Deaths in the United States, 1999-2015.

OBJECTIVES: A complete and accurate count of the number of opioid-related overdose deaths is essential to properly allocate resources. We determined the rate of unintentional overdose deaths (non-opioid-related, opioid-related, or unspecified) in the United States and by state from 1999 to 2015 and the possible effects of underreporting on national estimates of opioid abuse. METHODS: We abstracted unintentional drug overdose deaths ( International Classification of Diseases, 10th Revision, codes X40-X44) with contributory drug-specific T codes (T36.0-T50.9) from the Mortality Multiple Cause Micro-Data Files. We assumed that the proportion of unspecified overdose deaths that might be attributed to opioids would be the same as the proportion of opioid-related overdose deaths among all overdose deaths and calculated the number of deaths that could be reallocated as opioid-related for each state and year. We then added these reallocated deaths to the reported deaths to determine their potential effect on total opioid-related deaths. RESULTS: From 1999 to 2015, a total of 438 607 people died from unintentional drug overdoses. Opioid-related overdose deaths rose 401% (from 5868 to 29 383), non-opioid-related overdose deaths rose 150% (from 3005 to 7505), and unspecified overdose deaths rose 220% (from 2255 to 29 383). In 5 states (Alabama, Indiana, Louisiana, Mississippi, and Pennsylvania), more than 35% of unintentional overdose deaths were coded as unspecified. Our reallocation resulted in classifying more than 70 000 unspecified overdose deaths as potential additional opioid-related overdose deaths. CONCLUSIONS: States may be greatly underestimating the effect of opioid-related overdose deaths because of incomplete cause-of-death reporting, indicating that the current opioid overdose epidemic may be worse than it appears.

Improvement in racial disparities in years of life lost in the USA since 1990.

OBJECTIVE: To examine changes in cause-specific Years of Life Lost (YLL) by age, race, and sex group in the USA from 1990 to 2014. METHODS: 60 million death reports from the National Center for Health Statistics (NCHS) were categorized by age group, sex, race, and cause of death. YLL were calculated using age-specific life expectancies. Age groups were: infants <1, children 1-19, adults 20-64, and older adults 65+. RESULTS: Blacks have historically experienced more years of life lost than whites or other racial groups in the USA. In the year 1990 the YLL per 100,000 population was 21,103 for blacks, 14,160 for whites, and 7,417 for others. Between 1990 and 2014 overall YLL in the USA improved by 10%, but with marked variations in the rate of change across age, race, and sex groups. Blacks (all ages, both sexes) showed substantial improvement with a 28% reduction in YLL, compared to whites (all ages, both sexes) who showed a 4% reduction. Among blacks, improvements were seen in all age groups: reductions of 43%, 48%, 28%, and 25% among infants, children, adults, and older adults, respectively. Among whites, reductions of 33%, 44%, and 18% were seen in infants, children, and older adults, but there was a 6% increase in YLL among white adults. YLL increased by 18% in white adult females and declined 1% in white adult males. American Indian/Alaska Native women also had worsening in YLL, with an 8% increase. Asian Pacific Islanders consistently had the lowest YLL across all ages. Whites had a higher proportion of YLL due to overdose; blacks had a higher proportion due to homicide at younger ages and to heart disease at older ages. CONCLUSIONS: Race-based disparities in YLL in the USA since 1990 have narrowed considerably, largely as a result of improvements among blacks compared to whites. Adult white and American Indian / Alaskan Native females have experienced worsening YLL, while white males have experienced essentially no change. If recent trajectories continue, adult black/white disparities in YLL will continue to narrow.

Mortality Among Hardmetal Production Workers: Swedish Measurement Data and Exposure Assessment.

BACKGROUND: Mortality pattern was determined in a cohort of 16,999 white and blue-collar workers in the Swedish hardmetal industry. Exposure assessment for cobalt is presented. METHODS: A historical database (1970 to 2012) of personal and area measurements of cobalt, tungsten, and nickel in the Swedish hardmetal industry was created. Log linear and exponential modeling of cobalt concentrations based on time period, job, and site was performed, and cumulative and mean exposures were calculated. RESULTS: Some 37% of the personal cobalt measurements exceeded 0.02 mg/m, mostly for powder production, pressing, and shaping. The log linear regression showed statistical differences (P < 0.05) between sites, time periods, and jobs. Some 1.6% of the cobalt cumulative exposures for blue-collar workers exceeded 0.4 mg/m years. CONCLUSION: Low levels of cumulative and mean exposures were determined.

Chronic myelogenous leukemia in eastern Pennsylvania: an assessment of registry reporting.

BACKGROUND: Chronic myelogenous leukemia (CML) has been reportable to the Pennsylvania Cancer Registry (PCR) since the 1980s, but the completeness of reporting is unknown. This study assessed CML reporting in eastern Pennsylvania where a cluster of another myeloproliferative neoplasm was previously identified. METHODS: Cases were identified from 2 sources: 1) PCR case reports for residents of Carbon, Luzerne, or Schuylkill County with International Classification of Diseases for Oncology, Third Edition (ICD-O-3) codes 9875 (CML, BCR-ABL+), 9863 (CML, NOS), and 9860 (myeloid leukemia) and date of diagnosis 2001-2009, and 2) review of billing records at hematology practices. Participants were interviewed and their medical records were reviewed by board-certified hematologists. RESULTS: PCR reports included 99 cases coded 9875 or 9863 and 9 cases coded 9860; 2 additional cases were identified by review of billing records. Of the 110 identified cases, 93 were mailed consent forms, 23 consented, and 12 medical records were reviewed. Hematologists confirmed 11 of 12 reviewed cases as CML cases; all 11 confirmed cases were BCR/ABL positive, but only 1 was coded as positive (code 9875). CONCLUSIONS: Very few unreported CML cases were identified, suggesting relatively complete reporting to the PCR. Cases reviewed were accurately diagnosed, but ICD-0-3 coding often did not reflect BCR-ABL-positive tests. Cancer registry abstracters should look for these test results and code accordingly.

Comparison of Mortality Disparities in Central Appalachian Coal- and Non-Coal-Mining Counties.

OBJECTIVE: Determine whether select cause of death mortality disparities in four Appalachian regions is associated with coal mining or other factors. METHODS: We calculated direct age-adjusted mortality rates and associated 95% confidence intervals by sex and study group for each cause of death over 5-year time periods from 1960 to 2009 and compared mean demographic and socioeconomic values between study groups via two-sample t tests. RESULTS: Compared with non-coal-mining areas, we found higher rates of poverty in West Virginia and Virginia (VA) coal counties. All-cause mortality rates for males and females were higher in coal counties across all time periods. Virginia coal counties had statistically significant excesses for many causes of death. CONCLUSIONS: We found elevated mortality and poverty rates in coal-mining compared with non-coal-mining areas of West Virginia and VA. Future research should examine these findings in more detail at the individual level.

Long-term health experience of jet engine manufacturing workers: IV. A comparison of central nervous system cancer ascertainment using mortality and incidence data.

PURPOSE: To compare ascertainment of central nervous system (CNS) neoplasms with the use of mortality and incidence data as part of an occupational epidemiology study. METHODS: Deaths were identified by matching the cohort of 223,894 jet engine manufacturing employees to the U.S. Social Security Administration death files and the National Death Index. Incident cancer cases were identified by matching the cohort to 19 state cancer registries. RESULTS: We identified 718 cases overall: 59% by the use of both mortality and cancer incidence tracing; 24% by the use of only mortality tracing, and 17% by the use of only cancer incidence tracing. Compared with state cancer registries, death certificates missed 38% of the malignant, more than six times the benign and nearly 1.5 times the unspecified CNS cases. The positive predictive value of death certificates, with cancer registry as gold standard, was 6% for unspecified, 35% for benign, and 86% for malignant histologies. CONCLUSIONS: Death certificates seriously underascertained benign and unspecified CNS tumors; analyses determined with mortality data would not accurately capture the true extent of disease among the cohort. Most state cancer registries have only collected nonmalignant CNS tumor information since 2004, which currently limits the usefulness of state cancer registries as a source of nonmalignant CNS tumor identification. Underascertainment of CNS deaths could seriously affect interpretation of results, more so if examining nonmalignant CNS.

Pharmaceutical production workers and the risks of mortality from respiratory system cancer and lymphatic and hematopoietic tissue cancers.

OBJECTIVES: To evaluate further elevated mortality risks from respiratory system cancer (RSC) and lymphatic and hematopoietic tissue cancers (LHTC) in a cohort of 1466 male workers employed full-time in pharmaceutical production. METHODS: We computed standardized mortality ratios, and in nested case-control studies of RSC and LHTC, evaluated mortality risks by plant exposures with adjustment for potential confounding factors. RESULTS: Subjects with potential plant exposure had no elevated RSC risk and a statistically significant LHTC excess. The case-control study found many RSC risks reduced with adjustment for smoking, and LHTC risks increased with increasing levels of average exposure to dimethyl-formamide. CONCLUSIONS: RSC mortality risks decreased, and we found limited evidence that positive confounding by smoking may explain some remaining excess risks. For LHTC, increased mortality risks and exposure-response patterns in the case-control study may indicate a possible workplace association.

Underascertainment of deaths using social security records: a recommended solution to a little-known problem.

Complete and accurate ascertainment of vital status is of great importance in cohort studies. Recently, during the vital status ascertainment phase of an ongoing occupational mortality study, the authors discovered a potentially serious problem with use of the Pension Benefit Information Company's tracing service or any tracing that relies on records from the Social Security Administration (SSA) Death Master File to identify deaths. Their investigation revealed that a number of US states restrict the information in the SSA's Death Master File that is available to researchers and the public as a source of death information. As a result of these findings, the authors recommend a revised two-stage vital status tracing protocol. For stage I, data on all subjects for whom vital status is unconfirmed should be sent to the SSA. For stage II, information on all subjects to whom SSA assigned an unknown vital status as well as all subjects whom SSA identified as known decedents should be submitted to the National Death Index. This new protocol will enable researchers to maximize vital status ascertainment while containing costs associated with death identification.

Mortality patterns among workers in a US pharmaceutical production plant.

PURPOSE: To examine mortality among workers in a pharmaceutical production plant and to address community concerns about 1980 to 1990 increases in local county cancer mortality rates. METHODS: Subjects were 1999 workers with some full-time employment during the period between 1970 and 1996. We identified deaths through the year 2000 and reconstructed exposures to nine chemical agents with available exposure measurements. Data analyses included standardized mortality ratios (SMRs) and time trends in local cancer mortality rates. RESULTS: We observed deficits in deaths from all causes combined, all cancers combined, and most cause of death categories examined. Male workers with potential plant exposure had excesses in deaths from all lymphatic-hematopoietic tissue cancers (LHTC), in particular non-Hodgkin's lymphoma (NHL), and respiratory system cancers (RSC) that were larger among long-term workers, but the pattern of findings suggested the excesses were probably not related to occupational factors at the plant. The increase in local county cancer mortality rates was simply the upward cycle of a periodic trend that peaked in 1990 and returned to 1980 levels in 2000. CONCLUSIONS: With the possible exceptions of LHTC, in particular NHL, and RSC, this study provided no evidence of elevated total or cause-specific cohort mortality risks. It does not appear that plant factors played a role in the 1980 to 1990 increases in local county cancer mortality rates.

A 50-year historical cohort mortality study of workers in a pharmaceutical plant.

An historical cohort study was conducted of workers at a pharmaceutical manufacturing plant. The cohort mortality experience of workers ever employed at the plant over the period from 1950 to 1999 was examined. The 1958 workers accumulated 44,294 person-years of experience at the plant, and a total of 384 deaths were identified. Our findings from external comparisons based on standardized mortality ratios (SMRs) in the cohort provide no evidence of excess mortality risk from all causes combined (SMR=0.75), all cancers combined (SMR=0.96), or from certain other individual causes of death. No patterns of excess mortality risk were apparent after stratifying on age and sex or job classification. The mortality experience of this cohort was generally more favorable than that of the general population.