Predicting healthcare expenditure by multimorbidity groups

Prediction of pharmaceutical and non-pharmaceutical expenditures associated with Diabetes Mellitus type II based on clinical risk

OBJECTIVE

To assess the effectiveness of different machine learning models in estimating the pharmaceutical and non-pharmaceutical expenditures associated with Diabetes Mellitus type II diagnosis, based on the clinical risk index determined by the analysis of comorbidities.

MATERIALS AND METHODS

In this cross-sectional study, we have used data from 11,028 anonymized records of patients admitted to a high-complexity hospital in Bogota, Colombia between 2017-2019 with a primary diagnosis of Diabetes. These cases were classified according to Charlson's comorbidity index in several risk categories. The main variables analyzed in this study are hospitalization costs (which include pharmaceutical and non-pharmaceutical expenditures), age, gender, length of stay, medicines and services consumed, and comorbidities assessed by the Charlson's index. The model's dependent variable is expenditure (composed of pharmaceutical and non-pharmaceutical expenditures). Based on these variables, different machine learning models (Multivariate linear regression, Lasso model, and Neural Networks) were used to estimate the pharmaceutical and non-pharmaceutical expenditures associated with the clinical risk classification. To evaluate the performance of these models, different metrics were used: Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2).

RESULTS

The results indicate that the Neural Networks model performed better in terms of accuracy in predicting pharmaceutical and non-pharmaceutical expenditures considering the clinical risk based on Charlson's comorbidity index. A deeper understanding and experimentation with Neural Networks can improve these preliminary results, therefore we can also conclude that the main variables used and those that were proposed can be used as predictors for the medical expenditures of patients with diabetes type-II.

CONCLUSIONS

With the increase of technology elements and tools, it is possible to build models that allow decision-makers in hospitals to improve the resource planning process given the accuracy obtained with the different models tested.

Predicting healthcare expenditure based on Adjusted Morbidity Groups to implement a needs-based capitation financing system

BACKGROUND

Due to population aging, healthcare expenditure is projected to increase substantially in developed countries like Spain. However, prior research indicates that health status, not merely age, is a key driver of healthcare costs. This study analyzed data from over 1.25 million residents of Spain's Murcia region to develop a capitation-based healthcare financing model incorporating health status via Adjusted Morbidity Groups (AMGs). The goal was to simulate an equitable area-based healthcare budget allocation reflecting population needs.

METHODS

Using 2017 data on residents' age, sex, AMG designation, and individual healthcare costs, generalized linear models were built to predict healthcare expenditure based on health status indicators. Multiple link functions and distribution families were tested, with model selection guided by information criteria, residual analysis, and goodness-of-fit statistics. The selected model was used to estimate adjusted populations and simulate capitated budgets for the 9 healthcare districts in Murcia.

RESULTS

The gamma distribution with logarithmic link function provided the best model fit. Comparisons of predicted and actual average costs revealed underfunded and overfunded areas within Murcia. If implemented, the capitation model would decrease funding for most districts (up to 15.5%) while increasing it for two high-need areas, emphasizing allocation based on health status and standardized utilization rather than historical spending alone.

CONCLUSIONS

AMG-based capitated budgeting could improve equity in healthcare financing across regions in Spain. By explicitly incorporating multimorbidity burden into allocation formulas, resources can be reallocated towards areas with poorer overall population health. Further policy analysis and adjustment is needed before full-scale implementation of such need-based global budgets.

Racial differences in healthcare expenditures for prevalent multimorbidity combinations in the USA: a cross-sectional study

BACKGROUND

We aimed to model total charges for the most prevalent multimorbidity combinations in the USA and assess model accuracy across Asian/Pacific Islander, African American, Biracial, Caucasian, Hispanic, and Native American populations.

METHODS

We used Cerner HealthFacts data from 2016 to 2017 to model the cost of previously identified prevalent multimorbidity combinations among 38 major diagnostic categories for cohorts stratified by age (45-64 and 65 +). Examples of prevalent multimorbidity combinations include lipedema with hypertension or hypertension with diabetes. We applied generalized linear models (GLM) with gamma distribution and log link function to total charges for all cohorts and assessed model accuracy using residual analysis. In addition to 38 major diagnostic categories, our adjusted model incorporated demographic, BMI, hospital, and census division information.

RESULTS

The mean ages were 55 (45-64 cohort, N = 333,094) and 75 (65 + cohort, N = 327,260), respectively. We found actual total charges to be highest for African Americans (means $78,544 [45-64], $176,274 [65 +]) and lowest for Hispanics (means $29,597 [45-64], $66,911 [65 +]). African American race was strongly predictive of higher costs (p < 0.05 [45-64]; p < 0.05 [65 +]). Each total charge model had a good fit. With African American as the index race, only Asian/Pacific Islander and Biracial were non-significant in the 45-64 cohort and Biracial in the 65 + cohort. Mean residuals were lowest for Hispanics in both cohorts, highest in African Americans for the 45-64 cohort, and highest in Caucasians for the 65 + cohort. Model accuracy varied substantially by race when multimorbidity grouping was considered. For example, costs were markedly overestimated for 65 + Caucasians with multimorbidity combinations that included heart disease (e.g., hypertension + heart disease and lipidemia + hypertension + heart disease). Additionally, model residuals varied by age/obesity status. For instance, model estimates for Hispanic patients were highly underestimated for most multimorbidity combinations in the 65 + with obesity cohort compared with other age/obesity status groupings.

CONCLUSIONS

Our finding demonstrates the need for more robust models to ensure the healthcare system can better serve all populations. Future cost modeling efforts will likely benefit from factoring in multimorbidity type stratified by race/ethnicity and age/obesity status.

Living longer in declining health: Factors driving healthcare costs among older people

BACKGROUND

Developed countries are facing challenges in caring for people who are living longer but with a greater morbidity burden. Such people are likely to be regular users of healthcare.

OBJECTIVES

Our analytical aim is to identify factors that explain healthcare costs among: (1) people over 55 years old; (2) the top 5% and 1% high-cost users among this population; (3) those that transition into the top 5% and 1% from one year to the next; (4) those that appear in the top 5% and 1% over multiple years; and (5) those that remain in the top 5% and 1% over consecutive years.

METHODS

The data covered 2011 to 2017 and comprised 1,485,170 observations for a random sample of 224,249 people aged over 55 years in the Catalan region of Spain. We analysed each person's annual healthcare costs across all public healthcare settings related to their age, gender, socio-economic status (SES), whether or not and when they died, and morbidity status, through Adjusted Morbidity Groups.

RESULTS

After controlling for morbidity status, the oldest people did not have the highest costs and were less likely to be among the most costly patients. There was also only a modest impact on costs associated with SES and with dying. Healthcare costs were substantially higher for those with a neoplasm or four or more long term conditions (LTCs), costs rising with the complexity of their conditions. These morbidity indicators were also the most important factors associated with being and remaining in the top 5% or top 1% of costs.

CONCLUSION

Our results suggest that age and proximity to death are poor predictors of higher costs. Rather, healthcare costs are explained mainly by morbidity status, particularly whether someone has neoplasms or multiple LTCs. Morbidity measures should be included in future studies of healthcare costs.

Novel integration of governmental data sources using machine learning to identify super-utilization among U.S. counties

Background

Super-utilizers consume the greatest share of resource intensive healthcare (RIHC) and reducing their utilization remains a crucial challenge to healthcare systems in the United States (U.S.). The objective of this study was to predict RIHC among U.S. counties, using routinely collected data from the U.S. government, including information on consumer spending, offering an alternative method for identifying super-utilization among population units rather than individuals.

Methods

Cross-sectional data from 5 governmental sources in 2017 were used in a machine learning pipeline, where target-prediction features were selected and used in 4 distinct algorithms. Outcome metrics of RIHC utilization came from the American Hospital Association and included yearly: (1) emergency rooms visit, (2) inpatient days, and (3) hospital expenditures. Target-prediction features included: 149 demographic characteristics from the U.S. Census Bureau, 151 adult and child health characteristics from the Centers for Disease Control and Prevention, 151 community characteristics from the American Community Survey, and 571 consumer expenditures from the Bureau of Labor Statistics. SHAP analysis identified important target-prediction features for 3 RIHC outcome metrics.

Results

2475 counties with emergency rooms and 2491 counties with hospitals were included. The median yearly emergency room visits per capita was 0.450 [IQR:0.318, 0.618], the median inpatient days per capita was 0.368 [IQR: 0.176, 0.826], and the median hospital expenditures per capita was $2104 [IQR: $1299.93, 3362.97]. The coefficient of determination (R2), calculated on the test set, ranged between 0.267 and 0.447. Demographic and community characteristics were among the important predictors for all 3 RIHC outcome metrics.

Conclusions

Integrating diverse population characteristics from numerous governmental sources, we predicted 3-outcome metrics of RIHC among U.S. counties with good performance, offering a novel and actionable tool for identifying super-utilizer segments in the population. Wider integration of routinely collected data can be used to develop alternative methods for predicting RIHC among population units.

Analysis of hospital costs by morbidity group for patients with severe mental illness

OBJECTIVES

The goal of this study is to analyse hospital costs and length of stay of patients admitted to psychiatric units in hospitals in a European region of the Mediterranean Arc. The aim is to identify the effects of comorbidities and other variables in order to create an explanatory cost model.

METHODS

In order to carry out the study, the Ministry of Health was asked to provide data on access to the mental health facilities of all hospitals in the region. Among other questions, this database identifies the most important diagnostic variables related to admission, like comorbidities, age and gender. The method used, based on the Manning-Mullahy algorithm, was linear regression. The results were measured by the statistical significance of the independent variables to determine which of them were valid to explain the cost of hospitalization.

RESULTS

Psychiatric inpatients can be divided into three main groups (psychotic, organic and neurotic), which have statistically significant differences in costs. The independent variables that were statistically significant (p <.05) and their respective beta and confidence intervals were: psychotic group (19,833.0 ± 317.3), organic group (9,878.4 ± 276.6), neurotic group (11,060.1 ± 287.6), circulatory system diseases (19,170 ± 517.6), injuries and poisoning (21,101.6 ± 738.7), substance abuse (20,580.6 ± 514, 6) and readmission (19,150.9 ± 555.4).

CONCLUSIONS

Unlike most health services, access to psychiatric facilities does not correlate with comorbidities due to the specific nature of this specialization. Patients admitted to psychosis had higher costs and a higher number of average staysKEY MESSAGESThe highest average hospital expenditure occurred in patients admitted for psychotic disorders.Due to the particularities of psychiatry units and unlike other medical specialties, the number of comorbidities did not influence the number of hospital stays or hospital expenditure.Apart from the main diagnostic group, the variables that were useful to explain hospital expenditure were the presence of poisoning and injuries as comorbidity, diseases of circulatory system as comorbidity, history of substance abuse and readmission.

Utilization and out-of-pocket expenses of primary care among the multimorbid elderly in China: A two-part model with nationally representative data

Background

Multimorbidity has become an essential public health issue that threatens human health and leads to an increased disease burden. Primary care is the prevention and management of multimorbidity by providing continuous, comprehensive patient-centered services. Therefore, the study aimed to investigate the determinants of primary care utilization and out-of-pocket expenses (OOPE) among multimorbid elderly to promote rational utilization of primary care and reduce avoidable economic burdens.

Methods

The study used data from CHARLS 2015 and 2018, which included a total of 4,384 multimorbid elderly aged 60 and above. Guided by Grossman theory, determinants such as education, gender, marriage, household economy, and so on were included in this study. A two-part model was applied to evaluate primary care utilization and OOPE intensity in multimorbid populations. And the robustness testing was performed to verify research results.

Results

Primary care visits rate and OOPE indicated a decline from 2015 to 2018. Concerning primary outpatient care, the elderly who were female (OR = 1.51, P < 0.001), married (OR = 1.24, P < 0.05), living in rural areas (OR = 1.77, P < 0.001) and with poor self-rated health (OR = 2.23, P < 0.001) had a significantly higher probability of outpatient utilization, whereas those with middle school education (OR = 0.61, P < 0.001) and better household economy (OR = 0.96, P < 0.001) had a significantly less likelihood of using outpatient care. Rural patients (β = -0.72, P < 0.05) may have lower OOPE, while those with better household economy (β = 0.29, P < 0.05; β = 0.58, P < 0.05) and poor self-rated health (β = 0.62, P < 0.001) occurred higher OOPE. Regarding primary inpatient care, adults who were living in rural areas (OR = 1.48, P < 0.001), covered by Urban Employee Basic Medical Insurance (UEBMI) or Urban Rural Basic Medical Insurance (URBMI) (OR = 2.46, P < 0.001; OR = 1.81, P < 0.001) and with poor self-rated health (OR = 2.30, P < 0.001) had a significantly higher probability of using inpatient care, whereas individuals who were female (OR = 0.74, P < 0.001), with middle school education (OR = 0.40, P < 0.001) and better household economy (OR = 0.04, P < 0.001) had a significantly lower tendency to use inpatient care. Significantly, more OOPE occurred by individuals who were women (β = 0.18, P < 0.05) and with better household economy (β = 0.40, P < 0.001; β = 0.62, P < 0.001), whereas those who were covered by URBMI (β = -0.25, P < 0.05) and satisfied with their health (β = -0.21, P < 0.05) had less OOPE.

Conclusion

To prompt primary care visits and reduce economic burden among subgroups, more policy support is in need, such as tilting professional medical staff and funding to rural areas, enhancing awareness of disease prevention among vulnerable groups and so on.

Can diverse population characteristics be leveraged in a machine learning pipeline to predict resource intensive healthcare utilization among hospital service areas?

Background

Super-utilizers represent approximately 5% of the population in the United States (U.S.) and yet they are responsible for over 50% of healthcare expenditures. Using characteristics of hospital service areas (HSAs) to predict utilization of resource intensive healthcare (RIHC) may offer a novel and actionable tool for identifying super-utilizer segments in the population. Consumer expenditures may offer additional value in predicting RIHC beyond typical population characteristics alone.

Methods

Cross-sectional data from 2017 was extracted from 5 unique sources. The outcome was RIHC and included emergency room (ER) visits, inpatient days, and hospital expenditures, all expressed as log per capita. Candidate predictors from 4 broad groups were used, including demographics, adults and child health characteristics, community characteristics, and consumer expenditures. Candidate predictors were expressed as per capita or per capita percent and were aggregated from zip-codes to HSAs using weighed means. Machine learning approaches (Random Forrest, LASSO) selected important features from nearly 1,000 available candidate predictors and used them to generate 4 distinct models, including non-regularized and LASSO regression, random forest, and gradient boosting. Candidate predictors from the best performing models, for each outcome, were used as independent variables in multiple linear regression models. Relative contribution of variables from each candidate predictor group to regression model fit were calculated.

Results

The median ER visits per capita was 0.482 [IQR:0.351–0.646], the median inpatient days per capita was 0.395 [IQR:0.214–0.806], and the median hospital expenditures per capita was $2,302 [1$,544.70-$3,469.80]. Using 1,106 variables, the test-set coefficient of determination (R²) from the best performing models ranged between 0.184–0.782. The adjusted R² values from multiple linear regression models ranged from 0.311–0.8293. Relative contribution of consumer expenditures to model fit ranged from 23.4–33.6%.

Discussion

Machine learning models predicted RIHC among HSAs using diverse population data, including novel consumer expenditures and provides an innovative tool to predict population-based healthcare utilization and expenditures. Geographic variation in utilization and spending were identified.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-022-08154-4.

Variability in Healthcare Expenditure According to the Stratification of Adjusted Morbidity Groups in the Canary Islands (Spain)

Morbidity is the main item in the distribution of expenditure on healthcare services. The Adjusted Morbidity Group (AMG) measures comorbidity and complexity and classifies the patient into mutually exclusive clinical categories. The aim of this study is to analyse the variability of healthcare expenditure on users with similar scores classified by the AMG. Observational analytical and retrospective study. Population: 1,691,075 subjects, from Canary Islands (Spain), aged over 15 years with data from health cards, clinical history, Basic Minimum Specialised Healthcare Data Set, AMG, hospital agreements information system and Electronic Prescriptions. A descriptive, bivariant (ANOVA coefficient η²) and multivariant analysis was conducted. There is a correlation between the costs and the weight of AMG (rho = 0.678) and the prescribed active ingredients (rho = 0.689), which is smaller with age and does not exist with the other variables. As for the influence of the AMG morbidity group on the total costs of the patient, the coefficient η² (0.09) obtains a median effect in terms of the variability of expenditure, hence there is intra- and inter-group variability in the cost. In a first model created with all the variables and the cost, an explanatory power of 36.43% (R² = 0.3643) was obtained; a second model that uses solely active ingredients, AMG weight, being female and a pensioner obtained an explanatory power of 36.4%. There is room for improvement in terms of predicting the expenditure.

Health and Housing Energy Expenditures: A Two-Part Model Approach

Interest in the interaction between energy and health within the built environment has been increasing in recent years, in the context of sustainable development. However, in order to promote health and wellbeing across all ages it is necessary to have a better understanding of the association between health and energy at household level. This study contributes to this debate by addressing the case of Portugal using data from the Household Budget Survey (HBS) microdata database. A two-part model is applied to estimate health expenditures based on energy-related expenditures, as well as socioeconomic variables. Additional statistical methods are used to enhance the perception of relevant predictors for health expenditures. Our findings suggest that given the high significance and coefficient value, energy expenditure is a relevant explanatory variable for health expenditures. This result is further validated by a dominance analysis ranking. Moreover, the results show that health gains and medical cost reductions can be a key factor to consider on the assessment of the economic viability of energy efficiency projects in buildings. This is particularly relevant for the older and low-income segments of the population.

Systematic review on the instruments used for measuring the association of the level of multimorbidity and clinically important outcomes

OBJECTIVES

There are multiple instruments for measuring multimorbidity. The main objective of this systematic review was to provide a list of instruments that are suitable for use in studies aiming to measure the association of a specific outcome with different levels of multimorbidity as the main independent variable in community-dwelling individuals. The secondary objective was to provide details of the requirements, strengths and limitations of these instruments, and the chosen outcomes.

METHODS

We conducted the review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PROSPERO registration number: CRD42018105297). We searched MEDLINE, Embase and CINAHL electronic databases published in English and manually searched the Journal of Comorbidity between 1 January 2010 and 23 October 2020 inclusive. Studies also had to select adult patients from primary care or general population and had at least one specified outcome variable. Two authors screened the titles, abstracts and full texts independently. Disagreements were resolved with a third author. The modified Newcastle-Ottawa Scale was used for quality assessment.

RESULTS

Ninety-six studies were identified, with 69 of them rated to have a low risk of bias. In total, 33 unique instruments were described. Disease Count and weighted indices like Charlson Comorbidity Index were commonly used. Other approaches included pharmaceutical-based instruments. Disease Count was the common instrument used for measuring all three essential core outcomes of multimorbidity research: mortality, mental health and quality of life. There was a rise in the development of novel weighted indices by using prognostic models. The data obtained for measuring multimorbidity were from sources including medical records, patient self-reports and large administrative databases.

CONCLUSIONS

We listed the details of 33 instruments for measuring the level of multimorbidity as a resource for investigators interested in the measurement of multimorbidity for its association with or prediction of a specific outcome.

Data Envelopment Analysis Applications on Primary Health Care Using Exogenous Variables and Health Outcomes

A data envelopment analysis was used to evaluate the efficiency of 18 primary healthcare centres in a health district of the Valencian Community, Spain. Factor analysis was used as a first step in order to identify the most explanatory variables to be incorporated in the models. Included as variable inputs were the ratios of general practitioners, nurses, and costs; as output variables, those included were consultations, emergencies, avoidable hospitalisations, and prescription efficiency; as exogenous variables, those included were the percentage of population over 65 and a multimorbidity index. Confidence intervals were calculated using bootstrapping to correct possible biases. Efficient organisations within the set were identified, although the results depend on the models used and the introduction of exogenous variables. Pharmaceutical expenditure showed the greatest slack and room for improvement in its management. Data envelopment analysis allows an evaluation of efficiency that is focussed on achieving better results and a proper distribution and use of healthcare resources, although it needs the desired goals of the healthcare managers to be clearly identified, as the perspective of the analysis influences the results, as does including variables that measure the achievements and outcomes of the healthcare services.

The role of mental disease on the association between multimorbidity and medical expenditure

BACKGROUND

Multimorbidity is the presence of two or more chronic diseases and is associated with increased adverse outcomes, including hospitalization, mortality and frequency of use of medical institutions.

OBJECTIVE

This study aimed to describe multimorbidity patterns, determine whether multimorbidity was associated with high medical expenditure, and determine whether mental diseases had an interaction effect on this association.

METHODS

We conducted a claims data-based observational study. Data were obtained for 7526 individuals aged 0-75 years from a medical claims data set for Goto, Japan, over a 12-month period (2016-17). Annual medical expenditure was divided into quintiles; the fifth quintile represented high medical expenditure. Multimorbidity status was defined as the occurrence of two or more health conditions from 17 specified conditions. Odds ratios (OR) and 95% confidence intervals (CI) for high medical expenditure were calculated by number of comorbidities.

RESULTS

In total, 5423 (72.1%) participants had multimorbidity. Multimorbidity was significantly associated with high medical expenditure, even after adjustment for age, sex and income category (OR: 10.36, 95% CI: 7.57-14.19; P < 0.001). Mental diseases had a significant interaction effect on the association between multimorbidity and high medical expenditure (P = 0.001).

CONCLUSIONS

Multimorbidity is associated with high medical expenditure in Japan. Mental diseases may contribute to increased medical costs.