Incorporating machine learning and social determinants of health indicators into prospective risk adjustment for health plan payments

Seeing the random forest through the decision trees. Supporting learning health systems from histopathology with machine learning models: Challenges and opportunities

This paper discusses some overlooked challenges faced when working with machine learning models for histopathology and presents a novel opportunity to support "Learning Health Systems" with them. Initially, the authors elaborate on these challenges after separating them according to their mitigation strategies: those that need innovative approaches, time, or future technological capabilities and those that require a conceptual reappraisal from a critical perspective. Then, a novel opportunity to support "Learning Health Systems" by integrating hidden information extracted by ML models from digitalized histopathology slides with other healthcare big data is presented.

A machine learning model to predict therapeutic inertia in type 2 diabetes using electronic health record data

OBJECTIVE

To estimate the therapeutic inertia prevalence for patients with type 2 diabetes, develop and validate a machine learning model predicting therapeutic inertia, and determine the added predictive value of area-level social determinants of health (SDOH).

METHODS

This prognostic study with a retrospective cohort design used OneFlorida data (linked electronic health records (EHRs) from 1240 practices/clinics in Florida). The study cohort included adults (aged ≥ 18) with type 2 diabetes, HbA1C ≥ 7% (53 mmol/mol), ≥one ambulatory visit, and ≥one antihyperglycemic medication prescribed (excluded patients prescribed insulin before HbA1C). The outcome was therapeutic inertia, defined as absence of treatment intensification within six months after HbA1C ≥ 7% (53 mmol/mol). The predictors were patient, provider, and healthcare system factors. Machine learning methods included gradient boosting machines (GBM), random forests (RF), elastic net (EN), and least absolute shrinkage and selection operator (LASSO). The DeLong test compared the discriminative ability (represented by C-statistics) between models.

RESULTS

The cohort included 31,087 patients with type 2 diabetes (mean age = 58.89 (SD = 13.27) years, 50.50% male, 58.89% White). The therapeutic inertia prevalence was 39.80% among the 68,445 records. GBM outperformed (C-statistic from testing sample = 0.84, 95% CI = 0.83-0.84) RF (C-statistic = 0.80, 95% CI = 0.79-0.80), EN (C-statistic = 0.80, 95% CI = 0.80-0.81), and LASSO (C-statistic = 0.80, 95% CI = 0.80-0.81), p < 0.05. Area-level SDOH significantly increased the discriminative ability versus models without SDOH (C-statistic for GBM = 0.84, 95% CI = 0.84-0.85 vs. 0.84, 95% CI = 0.83-0.84), p < 0.05.

CONCLUSIONS

Using EHRs of patients with type 2 diabetes from a large state, machine learning predicted therapeutic inertia (prevalence = 40%). The model's ability to predict patients at high risk of therapeutic inertia is clinically applicable to diabetes care.

Updating the Chronic Illness and Disability Payment System

BACKGROUND

Of the 38 Medicaid programs that risk adjust payments to Medicaid managed care organizations (MCOs), 33 of them use the Chronic Illness and Disability Payment System (CDPS). There has been recent interest in adding social determinants of health (SDH) into risk-adjustment models.

OBJECTIVE

To update the CDPS models using recent MCO data based on the International Classification of Diseases version 10 coding system and to explore whether indicators of SDH are predictive of expenditures.

RESEARCH DESIGN

Data from 3 national Medicaid MCOs and 8 states are used to update the CDPS model. We test whether spending on Medicaid beneficiaries living in economically and socially deprived communities is greater than spending on similar beneficiaries in less deprived communities.

SUBJECTS

Medicaid beneficiaries with full benefits and without dual eligibility under Medicare enrolled in Medicaid MCOs in 8 states during 2017-2019, including 1.4M disabled beneficiaries, 9.2M children, and 6.4M adults.

MEASURES

Health care eligibility and claims records. Indicators based on the Social Deprivation Index were used to measure SDH.

RESULTS

The revised CDPS model has 52 CDPS categories within 19 major categories. Six major categories of CDPS were revised: Psychiatric, Pulmonary, Renal, Cancer, Infectious Disease, and Hematological. We found no relationship between health care spending and the Social Deprivation Index.

CONCLUSIONS

The revised CDPS models and regression weights reflect the updated International Classification of Diseases-10 coding system and recent managed care delivery. States should choose alternative payment strategies to address disparities in health and health outcomes.

Prediction of non emergent acute care utilization and cost among patients receiving Medicaid

Patients receiving Medicaid often experience social risk factors for poor health and limited access to primary care, leading to high utilization of emergency departments and hospitals (acute care) for non-emergent conditions. As programs proactively outreach Medicaid patients to offer primary care, they rely on risk models historically limited by poor-quality data. Following initiatives to improve data quality and collect data on social risk, we tested alternative widely-debated strategies to improve Medicaid risk models. Among a sample of 10 million patients receiving Medicaid from 26 states and Washington DC, the best-performing model tripled the probability of prospectively identifying at-risk patients versus a standard model (sensitivity 11.3% [95% CI 10.5, 12.1%] vs 3.4% [95% CI 3.0, 4.0%]), without increasing "false positives" that reduce efficiency of outreach (specificity 99.8% [95% CI 99.6, 99.9%] vs 99.5% [95% CI 99.4, 99.7%]), and with a ~ tenfold improved coefficient of determination when predicting costs (R2: 0.195-0.412 among population subgroups vs 0.022-0.050). Our best-performing model also reversed the lower sensitivity of risk prediction for Black versus White patients, a bias present in the standard cost-based model. Our results demonstrate a modeling approach to substantially improve risk prediction performance and equity for patients receiving Medicaid.

Risk adjustment for regional healthcare funding allocations with ensemble methods: an empirical study and interpretation

We experiment with recent ensemble machine learning methods in estimating healthcare costs, utilizing Finnish data containing rich individual-level information on healthcare costs, socioeconomic status and diagnostic data from multiple registries. Our data are a random 10% sample (553,675 observations) from the Finnish population in 2017. Using annual healthcare cost in 2017 as a response variable, we compare the performance of Random forest, Gradient Boosting Machine (GBM) and eXtreme Gradient Boosting (XGBoost) to linear regression. As machine learning methods are often seen as unsuitable in risk adjustment applications because of their relative opaqueness, we also introduce visualizations from the machine learning literature to help interpret the contribution of individual variables to the prediction. Our results show that ensemble machine learning methods can improve predictive performance, with all of them significantly outperforming linear regression, and that a certain level of interpretation can be provided for them. We also find individual-level socioeconomic variables to improve prediction accuracy and that their effect is larger for machine learning methods. However, we find that the predictions used for funding allocations are sensitive to model selection, highlighting the need for comprehensive robustness testing when estimating risk adjustment models used in applications.

Developing a Machine Learning Risk-adjustment Method for Hospitalizations and Emergency Department Visits of Nursing Home Residents With Dementia

BACKGROUND

Long-stay nursing home (NH) residents with Alzheimer disease and related dementias (ADRD) are at high risk of hospital transfers. Machine learning might improve risk-adjustment methods for NHs.

OBJECTIVES

The objective of this study was to develop and compare NH risk-adjusted rates of hospitalizations and emergency department (ED) visits among long-stay residents with ADRD using Extreme Gradient Boosting (XGBoost) and logistic regression.

RESEARCH DESIGN

Secondary analysis of national Medicare claims and NH assessment data in 2012 Q3. Data were equally split into the training and test sets. Both XGBoost and logistic regression predicted any hospitalization and ED visit using 58 predictors. NH-level risk-adjusted rates from XGBoost and logistic regression were constructed and compared. Multivariate regressions examined NH and market factors associated with rates of hospitalization and ED visits.

SUBJECTS

Long-stay Medicare residents with ADRD (N=413,557) from 14,057 NHs.

RESULTS

A total of 8.1% and 8.9% residents experienced any hospitalization and ED visit in a quarter, respectively. XGBoost slightly outperformed logistic regression in area under the curve (0.88 vs. 0.86 for hospitalization; 0.85 vs. 0.83 for ED visit). NH-level risk-adjusted rates from XGBoost were slightly lower than logistic regression (hospitalization=8.3% and 8.4%; ED=8.9% and 9.0%, respectively), but were highly correlated. Facility and market factors associated with the XGBoost and logistic regression-adjusted hospitalization and ED rates were similar. NHs serving more residents with ADRD and having a higher registered nurse-to-total nursing staff ratio had lower rates.

CONCLUSIONS

XGBoost and logistic regression provide comparable estimates of risk-adjusted hospitalization and ED rates.

Using machine-learning algorithms to improve imputation in the medical expenditure panel survey

OBJECTIVE

To assess the feasibility of applying machine learning (ML) methods to imputation in the Medical Expenditure Panel Survey (MEPS).

DATA SOURCES

All data come from the 2016-2017 MEPS.

STUDY DESIGN

Currently, expenditures for medical encounters in the MEPS are imputed with a predictive mean matching (PMM) algorithm in which a linear regression model is used to predict expenditures for events with (donors) and without (recipients) data. Recipient events and donor events are then matched based on the smallest distance between predicted expenditures, and the donor event's expenditures are used as the recipient event's imputation. We replace linear regression algorithm in the PMM framework with ML methods to predict expenditures. We examine five alternatives to linear regression: Gradient Boosting, Random Forests, Extreme Random Forests, Deep Neural Networks, and a Stacked Ensemble approach. Additionally, we introduce an alternative matching scheme, which matches on a vector of predicted expenditures by sources of payment instead of a single total expenditure prediction to generate potentially superior matches.

DATA COLLECTION

Study data is derived from a large federal survey.

PRINCIPAL FINDINGS

ML algorithms perform better at both prediction and matching imputation than Ordinary Least Squares (OLS), the most common prediction algorithm used in PMM. On average, the Stacked Ensemble approach that combines all the ML algorithms performs best, improving expenditure prediction R2 by 108% (0.156 points) and final imputation R2 by 227% (0.397 points). Matching on a prediction vector also improves alignment of sources of payments between donor and recipient events.

CONCLUSIONS

ML algorithms and an alternative matching scheme improve the overall quality of expenditure PMM imputation in the MEPS. These methods may have additional value in other national surveys that currently rely on PMM or similar methods for imputation.

Social determinants of health in prognostic machine learning models for orthopaedic outcomes: A systematic review

RATIONAL

Social determinants of health (SDOH) are being considered more frequently when providing orthopaedic care due to their impact on treatment outcomes. Simultaneously, prognostic machine learning (ML) models that facilitate clinical decision making have become popular tools in the field of orthopaedic surgery. When ML-driven tools are developed, it is important that the perpetuation of potential disparities is minimized. One approach is to consider SDOH during model development. To date, it remains unclear whether and how existing prognostic ML models for orthopaedic outcomes consider SDOH variables.

OBJECTIVE

To investigate whether prognostic ML models for orthopaedic surgery outcomes account for SDOH, and to what extent SDOH variables are included in the final models.

METHODS

A systematic search was conducted in PubMed, Embase and Cochrane for studies published up to 17 November 2020. Two reviewers independently extracted SDOH features using the PROGRESS+ framework (place of residence, race/ethnicity, Occupation, gender/sex, religion, education, social capital, socioeconomic status, 'Plus+' age, disability, and sexual orientation).

RESULTS

The search yielded 7138 studies, of which 59 met the inclusion criteria. Across all studies, 96% (57/59) considered at least one PROGRESS+ factor during development. The most common factors were age (95%; 56/59) and gender/sex (96%; 57/59). Differential effect analyses, such as subgroup analysis, covariate adjustment, and baseline comparison, were rarely reported (10%; 6/59). The majority of models included age (92%; 54/59) and gender/sex (69%; 41/59) as final input variables. However, factors such as insurance status (7%; 4/59), marital status (7%; 4/59) and income (3%; 2/59) were seldom included.

CONCLUSION

The current level of reporting and consideration of SDOH during the development of prognostic ML models for orthopaedic outcomes is limited. Healthcare providers should be critical of the models they consider using and knowledgeable regarding the quality of model development, such as adherence to recognized methodological standards. Future efforts should aim to avoid bias and disparities when developing ML-driven applications for orthopaedics.

Risk Adjustment And Promoting Health Equity In Population-Based Payment: Concepts And Evidence

The objective of risk adjustment is not to predict spending accurately but to support the social goals of a payment system, which include equity. Setting population-based payments at accurate predictions risks entrenching spending levels that are insufficient to mitigate the impact of social determinants on health care use and effectiveness. Instead, to advance equity, payments must be set above current levels of spending for historically disadvantaged groups. In analyses intended to guide such reallocations, we found that current risk adjustment for the community-dwelling Medicare population overpredicts annual spending for Black and Hispanic beneficiaries by $376-$1,264. The risk-adjusted spending for these populations is lower than spending for White beneficiaries despite the former populations' worse risk-adjusted health and functional status. Thus, continued movement from fee-for-service to population-based payment models that omit race and ethnicity from risk adjustment (as current models do) should result in sizable resource reallocations and incentives that support efforts to address racial and ethnic disparities in care. We found smaller overpredictions for less-educated beneficiaries and communities with higher proportions of residents who are Black, Hispanic, or less educated, suggesting that additional payment adjustments that depart from predictive accuracy are needed to support health equity. These findings also suggest that adding social risk factors as predictors to spending models used for risk adjustment may be counterproductive or accomplish little.

Using machine learning to study the effect of medication adherence in Opioid Use Disorder

BACKGROUND

Opioid Use Disorder (OUD) and opioid overdose (OD) impose huge social and economic burdens on society and health care systems. Research suggests that Medication for Opioid Use Disorder (MOUD) is effective in the treatment of OUD. We use machine learning to investigate the association between patient's adherence to prescribed MOUD along with other risk factors in patients diagnosed with OUD and potential OD following the treatment.

METHODS

We used longitudinal Medicaid claims for two selected US states to subset a total of 26,685 patients with OUD diagnosis and appropriate Medicaid coverage between 2015 and 2018. We considered patient age, sex, region level socio-economic data, past comorbidities, MOUD prescription type and other selected prescribed medications along with the Proportion of Days Covered (PDC) as a proxy for adherence to MOUD as predictive variables for our model, and overdose events as the dependent variable. We applied four different machine learning classifiers and compared their performance, focusing on the importance and effect of PDC as a variable. We also calculated results based on risk stratification, where our models separate high risk individuals from low risk, to assess usefulness in clinical decision-making.

RESULTS

Among the selected classifiers, the XGBoost classifier has the highest AUC (0.77) closely followed by the Logistic Regression (LR). The LR has the best stratification result: patients in the top 10% of risk scores account for 35.37% of overdose events over the next 12 month observation period. PDC score calculated over the treatment window is one of the most important features, with better PDC lowering risk of OD, as expected. In terms of risk stratification results, of the 35.37% of overdose events that the predictive model could detect within the top 10% of risk scores, 72.3% of these cases were non-adherent in terms of their medication (PDC <0.8). Targeting the top 10% outcome of the predictive model could decrease the total number of OD events by 10.4%.

CONCLUSIONS

The best performing models allow identification of, and focus on, those at high risk of opioid overdose. With MOUD being included for the first time as a factor of interest, and being identified as a significant factor, outreach activities related to MOUD can be targeted at those at highest risk.

Social Determinants of Health Influence Future Health Care Costs in the Medicaid Cohort of the District of Columbia Study

Policy Points Social determinants of health are an important predictor of future health care costs. Medicaid must partner with other sectors to address the underlying causes of its beneficiaries' poor health and high health care spending.

CONTEXT

Social determinants of health are an important predictor of future health care costs but little is known about their impact on Medicaid spending. This study analyzes the role of social determinants of health (SDH) in predicting future health care costs for adult Medicaid beneficiaries with similar past morbidity burdens and past costs.

METHODS

We enrolled into a prospective cohort study 8,892 adult Medicaid beneficiaries who presented for treatment at an emergency department or clinic affiliated with two hospitals in Washington, DC, between September 2017 and December 31, 2018. We used SDH information measured at enrollment to categorize our participants into four social risk classes of increasing severity. We used Medicaid claims for a 2-year period; 12 months pre- and post-study enrollment to measure past and future morbidity burden according to the Adjusted Clinical Groups system. We also used the Medicaid claims data to characterize total annual Medicaid costs one year prior to and one year after study enrollment.

RESULTS

The 8,892 participants were primarily female (66%) and Black (91%). For persons with similar past morbidity burdens and past costs (p < 0.01), the future morbidity burden was significantly higher in the upper two social risk classes (1.15 and 2.04, respectively) compared with the lowest one. Mean future health care spending was significantly higher in the upper social risk classes compared with the lowest one ($2,713, $11,010, and $17,710, respectively) and remained significantly higher for the two highest social risk classes ($1,426 and $3,581, respectively), given past morbidity burden and past costs (p < 0.01). When we controlled for future morbidity burden (measured concurrently with future costs), social risk class was no longer a significant predictor of future health care costs.

CONCLUSIONS

SDH are statistically significant predictors of future morbidity burden and future costs controlling for past morbidity burden and past costs. Further research is needed to determine whether current payment systems adequately account for differences in the care needs of highly medically and socially complex patients.

Estimation and Prediction of Hospitalization and Medical Care Costs Using Regression in Machine Learning

Medical costs are one of the most common recurring expenses in a person’s life. Based on different research studies, BMI, ageing, smoking, and other factors are all related to greater personal medical care costs. The estimates of the expenditures of health care related to obesity are needed to help create cost-effective obesity prevention strategies. Obesity prevention at a young age is a top concern in global health, clinical practice, and public health. To avoid these restrictions, genetic variants are employed as instrumental variables in this research. Using statistics from public huge datasets, the impact of body mass index (BMI) on overall healthcare expenses is predicted. A multiview learning architecture can be used to leverage BMI information in records, including diagnostic texts, diagnostic IDs, and patient traits. A hierarchy perception structure was suggested to choose significant words, health checks, and diagnoses for training phase informative data representations, because various words, diagnoses, and previous health care have varying significance for expense calculation. In this system model, linear regression analysis, naive Bayes classifier, and random forest algorithms were compared using a business analytic method that applied statistical and machine-learning approaches. According to the results of our forecasting method, linear regression has the maximum accuracy of 97.89 percent in forecasting overall healthcare costs. In terms of financial statistics, our methodology provides a predictive method.

Mortality Prediction Following Cardiac Surgery in Children - An STS Congenital Heart Surgery Database Analysis

BACKGROUND

The Society of Thoracic Surgeons' Congenital Heart Surgery Database (STS CHSD) provides risk-adjusted operative mortality rates to approximately 120 North American congenital heart centers. Optimal case-mix adjustment methods for operative mortality risk prediction in this population remain unclear.

METHODS

A panel created diagnosis-procedure (D-P) combinations of encounters in the CHSD. Models for operative mortality using the new D-P categories, procedure-specific risk factors, and syndromes/abnormalities included in the CHSD were estimated using Bayesian additive regression trees (BART) and lasso models. Performance of the new models was compared to the current STS-CHSD risk model.

RESULTS

Of 98,825 operative encounters (69,063 training; 29,762 testing), 2,818 (2.85%) STS-defined operative mortalities were observed. Differences in sensitivity, specificity, true and false positive predicted values were negligible across models. Calibration for mortality predictions at the higher end of risk from the lasso and BART models was better than predictions from the STS-CHSD model, likely due to new models' inclusion of diagnosis-palliative procedure variables affecting < 1% of patients overall, but accounting for 27% of mortalities. Model discrimination varied across models for high-risk procedures, hospital volume, and hospitals.

CONCLUSIONS

Overall performance of the new models did not differ meaningfully from the STS-CHSD risk model. Addition of procedure-specific risk factors and allowing diagnosis to modify predicted risk for palliative operations may augment model performance for very high-risk surgeries. Given the importance of risk adjustment in estimating hospital quality, a comparative assessment of surgical program quality evaluations using the different models is warranted.

Association Between Neighborhood-Level Social Determinants of Health and Access to Pediatric Appendicitis Care

IMPORTANCE

Presenting with complicated appendicitis, which is associated with higher rates of complications and readmissions compared with simple appendicitis, may indicate delayed access to care. Although both patient-level and neighborhood-level social determinants of health are associated with access to care, little is known about the association between neighborhood factors and access to acute pediatric surgical care.

OBJECTIVE

To examine the association between neighborhood factors and the odds of presenting with complicated appendicitis and unplanned postdischarge health care use.

DESIGN, SETTING, AND PARTICIPANTS

A retrospective cohort study of patients aged 18 years or younger diagnosed with appendicitis was conducted. Discharge data from October 1, 2015, to September 30, 2018, were obtained from the Pediatric Health Information System Database and linked to the Child Opportunity Index (COI) 2.0 Database. Data analysis was conducted from January 1 through July 1, 2021.

EXPOSURES

The COI, a composite score of zip code neighborhood opportunity level information, divided into quintiles ranging from very low to very high opportunity.

MAIN OUTCOMES AND MEASURES

Based on COI level, the main outcome was the odds of presenting with complicated appendicitis, which was defined using the Agency for Healthcare Research and Quality-specified International Statistical Classification of Diseases, 10th Edition, Clinical Modification codes. The secondary outcome was the odds of unplanned postdischarge health care use (emergency department visits and/or readmissions) for patients with simple and with complicated appendicitis.

RESULTS

A total of 67 489 patients (mean [SD] age, 10.5 [3.9] years) had appendicitis, with 31 223 cases (46.3%) being complicated. A total of 1699 patients (2.5%) were Asian, 24 234 (35.9%) were Hispanic, 4447 (6.6%) were non-Hispanic Black, and 29 234 (43.3%) were non-Hispanic White; 40 549 patients (60.1%) were male; and 32 343 (47.9%) were publicly insured. Patients living in very low-COI neighborhoods had 28% higher odds of presenting with complicated appendicitis (odds ratio, 1.28; 95% CI, 1.20-1.35) compared with those in very high-COI neighborhoods. There was no significant association between COI level and unplanned postdischarge health care use (very high COI, 20.8%; very low COI, 19.1%).

CONCLUSIONS AND RELEVANCE

In this cohort study, children from lower-COI neighborhoods had increased odds of presenting with complicated appendicitis compared with those from higher-COI neighborhoods, even after controlling for patient-level social determinants of health factors. These findings may inform policies and programs that seek to improve access to pediatric surgical care.

Explaining the Variance in Cardiovascular Disease Risk Factors: A Comparison of Demographic, Socioeconomic, and Genetic Predictors

BACKGROUND

Efforts to explain the burden of cardiovascular disease (CVD) often focus on genetic factors or social determinants of health. There is little evidence on the comparative predictive value of each, which could guide clinical and public health investments in measuring genetic versus social information. We compared the variance in CVD-related outcomes explained by genetic versus socioeconomic predictors.

METHODS

Data were drawn from the Health and Retirement Study (N = 8,720). We examined self-reported diabetes, heart disease, depression, smoking, and body mass index, and objectively measured total and high-density lipoprotein cholesterol. For each outcome, we compared the variance explained by demographic characteristics, socioeconomic position (SEP), and genetic characteristics including a polygenic score for each outcome and principal components (PCs) for genetic ancestry. We used R-squared values derived from race-stratified multivariable linear regressions to evaluate the variance explained.

RESULTS

The variance explained by models including all predictors ranged from 3.7% to 14.3%. Demographic characteristics explained more than half this variance for most outcomes. SEP explained comparable or greater variance relative to the combination of the polygenic score and PCs for most conditions among both white and Black participants. The combination of SEP, polygenic score, and PCs performed substantially better, suggesting that each set of characteristics may independently contribute to the prediction of CVD-related outcomes. Philip R. Lee Institute for Health Policy Studies, Department of Family & Community Medicine, UCSF.

CONCLUSIONS

Focusing on genetic inputs into personalized medicine predictive models, without considering measures of social context that have clear predictive value, needlessly ignores relevant information that is more feasible and affordable to collect on patients in clinical settings. See video abstract at, http://links.lww.com/EDE/B879.

Social Determinants of Health in Physiatry: Challenges and Opportunities for Clinical Decision Making and Improving Treatment Precision

Physiatry is a medical specialty focused on improving functional outcomes in patients with a variety of medical conditions that affect the brain, spinal cord, peripheral nerves, muscles, bones, joints, ligaments, and tendons. Social determinants of health (SDH) play a key role in determining therapeutic process and patient functional outcomes. Big data and precision medicine have been used in other fields and to some extent in physiatry to predict patient outcomes, however many challenges remain. The interplay between SDH and physiatry outcomes is highly variable depending on different phases of care, and more favorable patient profiles in acute care may be less favorable in the outpatient setting. Furthermore, SDH influence which treatments or interventional procedures are accessible to the patient and thus determine outcomes. This opinion paper describes utility of existing datasets in combination with novel data such as movement, gait patterning and patient perceived outcomes could be analyzed with artificial intelligence methods to determine the best treatment plan for individual patients in order to achieve maximal functional capacity.

Gender difference in under-reporting hiring discrimination in South Korea: a machine learning approach

OBJECTIVES

This study was conducted to examine gender differences in under-reporting hiring discrimination by building a prediction model for workers who responded “not applicable (NA)” to a question about hiring discrimination despite being eligible to answer.

METHODS

Using data from 3,576 wage workers in the seventh wave (2004) of the Korea Labor and Income Panel Study, we trained and tested 9 machine learning algorithms using “yes” or “no” responses regarding the lifetime experience of hiring discrimination. We then applied the best-performing model to estimate the prevalence of experiencing hiring discrimination among those who answered “NA.” Under-reporting of hiring discrimination was calculated by comparing the prevalence of hiring discrimination between the “yes” or “no” group and the “NA” group.

RESULTS

Based on the predictions from the random forest model, we found that 58.8% of the “NA” group were predicted to have experienced hiring discrimination, while 19.7% of the “yes” or “no” group reported hiring discrimination. Among the “NA” group, the predicted prevalence of hiring discrimination for men and women was 45.3% and 84.8%, respectively.

CONCLUSIONS

This study introduces a methodological strategy for epidemiologic studies to address the under-reporting of discrimination by applying machine learning algorithms.