Low-Value Care and Excess Out-Of-Pocket Expenditure Among Older Adults with Incident Cancer – A Machine learning approach.

Abstract

Objective:

To evaluate the association of low-value care with excess out-of-pocket expenditure among older adults diagnosed with incident breast, prostate, colorectal cancers, and Non-Hodgkin’s Lymphoma.

Methods:

We used a retrospective cohort study design with 12-month baseline and follow-up periods. We identified a cohort of older adults (age ≥ 66 years) diagnosed with breast, prostate, colorectal cancers, or Non-Hodgkin’s lymphoma between January 2014 and December 2014. We assessed low-value care and patient out-of-pocket expenditure in the follow-up period. We identified relevant low-value services using ICD9/ICD10 and CPT/HCPCS codes from the linked health claims and patient out-of-pocket expenditure from Medicare claim files and expressed expenditure in 2016 USD.

Results:

About 29% of older adults received at least one low-value care procedure during the follow-up period. Low-value care differed by gender, and rates were higher in women with colorectal cancer (32.7%) vs. (28.8%) and NHL (40%) vs. (39%) compared to men. Individuals who received one or more low-value care had significantly higher mean out-of-pocket expenditure ($8,726±$7,214) vs. ($6,802±$6,102). XGBOOST, a machine learning algorithm revealed that low-value care was among the five leading predictors of OOP expenditure.

Conclusion:

One in four older adults with incident cancer received low-value care in 12-months after a cancer diagnosis. Across all cancer populations, individuals who received low-value care had significantly higher out-of-pocket expenditure. Excess out-of-pocket expenditure was driven by low-value care, fragmentation of care, and an increasing number of pre-existing chronic conditions.

Policy Statement:

This study focuses on health policy issues, specifically value-based care and its findings have important clinical and policy implications for Centers for Medicare and Medicaid Services (CMS) which has issued a roadmap for states to accelerate the adoption of value-based care, with the Department of Health and Human Services (HHS) setting a goal of converting 50% of traditional Medicare payment systems to alternative payment models tied to value-based care by 2022.

Introduction

In the United States (U.S.), over 25% ($955 billion) of annual healthcare spending is considered wasteful because they are spent reimbursing low-value care.^1,2 Low-value care refers to the provision of healthcare services, medical tests, and procedures that have unclear or no clinical benefit to patients.³ Low-value care includes both overuse and overtreatment – defined as inappropriate care or one that may be inappropriate only under certain circumstances.⁴ Worldwide, low-value care imposes a significant economic burden on patients, payers, governments, and society.⁵ Previous studies have reported the excessive economic burden of low-value care on payers and government,^6–9 but there is limited literature on the economic burden to patients, specifically the out-of-pocket spending associated with low-value care. In Minnesota, 18 low-value care services accounted for $54.9 million spent in 2014, and 16.9% ($9.3 million) of that expenditure was borne out-of-pocket by patients.¹⁰ Kao-ping et al. evaluated 20 pediatric low-value services and reported that 33.9% ($9.3 million) of the overall expenditure was paid out-of-pocket.¹¹ No previous studies have estimated the economic burden of low-value care among cancer patients > 65 years.

Cancer care among older adults > 65 years is one of the biggest drivers of healthcare expenditure in the U.S. and accounts for nearly 40% of all spending.¹² Low-value care among cancer patients contributes to the financial toxicity of cancer treatment.¹³ However, most studies on the financial toxicity of low-value care among cancer patients focus on expensive cancer therapy, and have not explored the toxicity associated with non-cancer low-value care received for pre-existing chronic conditions. Examining the relationship between low-value care and out-of-pocket expenditure among cancer patients will provide meaningful data for policymakers on the economic impact of low-value care in elderly Medicare beneficiaries and help drive efforts to improve quality. This study aims to evaluate the association of low-value care to out-of-pocket expenditure among patients diagnosed with incident breast, prostate, colorectal cancers, and Non-Hodgkin’s Lymphoma.

Methods

Data Source:

We used data from multiple sources for this study. We combined data from the Surveillance, Epidemiology, and End Results (SEER) registry, Medicare claims (inpatient, outpatient, physician services, and home health agency claims), Area Health Resources File (AHRF), census data, and 5% non-cancer sample from the SEER regions. The SEER Program is a population-based tumor registry that collects data on all incident cases of cancer in persons residing in the 18 SEER regions. SEER cancer registry regions include (Atlanta, Connecticut, Detroit, Hawaii, Iowa, New Mexico, San Francisco-Oakland, Seattle-Puget Sound, Utah, Los Angeles and San Jose-Monterey, Rural Georgia, Alaska Native, Greater California, Kentucky, Louisiana, New Jersey, and Greater Georgia). Medicare is a federal health insurance program introduced by the Medicare bill of 1965 to serve as the primary health insurance program for the elderly. As of December 2020, it was estimated that more than 90% of adults over 65 years (52.7 million beneficiaries) received health insurance through Medicare.¹⁴ The original Medicare bill contained two main parts; A and B. Part A of the Medicare program covers services like inpatient hospitalization, Skilled Nursing Facility (SNF), home health services following hospital stay (HHA) and hospice care. Part B covers physician services, outpatient services, diagnostic tests, Durable Medical Equipment (DME), emergency room visits, home health care not following a hospital stay, laboratory services, and other medical services. In 2006, the Medicare plan was expanded to include an optional prescription drug benefit labeled Medicare Part D. An optional Medicare part C, also called Medicare Advantage was introduced with the Tax Equity and Fiscal Responsibility Act of 1982 which authorized Medicare to contract with risk-based private health plans that agree to assume liability for beneficiaries’ health expenses in exchange for a prospective monthly capitation fee for each covered person. Various forms of Medicare Advantage plans exist but the most common types are Health Maintenance Organizations (HMO) plans, Preferred Provider Organization (PPO) plans, and private Fee-for-Service (FFS) plans. Details concerning Medicare cost sharing, deductibles, coinsurance, copays and patient responsibility are given elsewhere.^14,15

Study Design:

We adopted a retrospective cohort study design with 12-month baseline and follow-up periods for this study. The index date was anchored to the date of cancer diagnosis, and the 12-month pre-diagnosis period was used as the baseline. We measured all the independent variables during the baseline period. We assessed low-value care and out-of-pocket expenditure in the 12-month follow-up period. The study cohort consisted of older adults (age ≥ 66 years) diagnosed with breast, prostate, colorectal cancers, or Non-Hodgkin’s lymphoma between January 2014 and December 2014. To be included in this study, Medicare beneficiaries were required to have one primary cancer diagnosis, be alive during the calendar year, and the cancer was not diagnosed from an autopsy report or death certificate. We included fee-for-service beneficiaries with continuous Part A and B enrollment during the study period and without HMO plan enrollment in the baseline and follow-up periods. We excluded individuals who were enrolled in HMO plans because Medicare data does not fully capture all claims for beneficiaries enrolled under these plans since the HMO’s are not required to submit all their claims to the Center for Medicare and Medicaid (CMS).

Measures

Target Dependent Variable

Out-of-pocket expenditure:

We identified all payments made by the patients for services obtained. Out-of-pocket payments included patient deductible, coinsurance, and blood deductible amounts. We assessed low-value care-specific out-of-pocket expenditure as well as the total out-of-pocket expenditure. We measured the total out-of-pocket expenditure for the cohort by adding the patient deductible amounts with patient co-payments and any other payments covered by the beneficiary using the already validated methodology described by ResDAC.¹⁶ We obtained all payment amounts from Medicare inpatient, outpatient, emergency room, home health agency, and durable medical equipment files. All healthcare expenditure was converted to real dollars using the consumer price index for medical services and expressed in 2016 USD.

Key Feature:

Low-value care:

Low-value care services were identified using previously published methodology by Schwartz et al.¹⁷ and Colla et al.⁷ representing some of the published Choosing Wisely recommendations. For each indicator procedure, we identified instances where the use of the procedure was most likely to be “low-value” because they were not clinically indicated. We used ICD9/ICD10 and CPT/HCPCS codes to identify relevant procedures from the inpatient, outpatient, physician, and home health agency files.

Other Features:

The selection of features was guided by published literature and the WHO cost of illness framework¹⁸ The framework assumes that healthcare utilization provides a level of utility to the consumer and leads to healthcare expenditure. The model considers both clinical and social determinants using a bottom-up approach to estimate expenditures and regards expenditures as the outcome of the healthcare process. Social determinants include factors like demographics (age at index date [66–69, 70–74, 75–79, or >=80 years], sex [male or female], race [non-Hispanic White, non-Hispanic African American, Latinx, or others]), geography (Northeast, South, North-Central, or West; rural/urban -metro, urban, or rural), socio-economic and access factors (zip-code level income and education, county-level Medicare and Medicaid dual eligibility), access to care (30-day readmission rates, emergency department visits per 1000 population, 3-year malignant neoplasm mortality rates, Medicare advantage penetration). Clinical factors include health status (chronic conditions including asthma, arthritis, chronic obstructive pulmonary disease, diabetes, coronary artery disease, congestive heart failure, hypertension, hyperlipidemia, chronic kidney disease; mental health conditions included dementia, bipolar disorder, depression, or anxiety), patient-level fragmentation of care, preventive services use and healthcare utilization. Fragmentation of care was measured using a claims-based Fragmentation of Care Index (FCI).¹⁹ FCI measures the dispersion of care across multiple providers and specialties; preventive service use consisted of receipt of influenza vaccination, and healthcare utilization was measured with E.D. and inpatient visits.

Analysis

Statistical analysis methods have historically been used to examine the relationship between risk factors (independent variables) and outcomes using a small number of clinically important variables and produce “clinician-friendly” measures of association like risk or odds ratios that are easy to interpret. These statistical models have some limitations related to strict assumptions about the data’s distribution and shape, including proportional hazards, additivity, and probability distribution of errors. These assumptions are often unmet in clinical practice, and violating these assumptions often renders the model inaccurate.

Machine Learning (ML) methods offer a flexible and adaptive alternative to statistical modeling because they do not rely on distributional assumptions. ML algorithms typically use recursive partitioning techniques to split original data into training, validation, and test datasets. These algorithms learn associations in the data using the training set and predict the “unseen” test data.

We built two ML models to assess the predictors of out-of-pocket expenditure; the first model was built using XGBOOST regression with log-transformed expenditure, while the second model was built using XGBOOST Tweedie regression using non-transformed expenditure. The Tweedie distributions are a special case of exponential distributions with “point mass” at zero and are helpful for modeling expenditure in datasets where most observations have zero expenditure.²⁰ We evaluated the performance of our models using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). To interpret our models, we use the Shapely Additive explanations (SHAP) explanation technique developed by Lundberg et al.²¹ SHAP values assess the importance of a feature by comparing the model prediction when the feature is present and when it is removed from the model. We also used SHAP accumulated local effect plots and partial dependence plots to explore the marginal effects of the top predictors of out-of-pocket expenditure.

Results

The study cohort comprised 27,067 eligible older adults who met all the inclusion criteria. About 29% of the individuals received at least one low-value care procedure in the follow-up period. A description of the cohort characteristics for adults with and without low-value care is presented in (Table 1). The majority of the cohort were female (52%), white (83%), lived in metro areas (84%), and had three or more pre-existing chronic conditions (59%). The most common chronic conditions include hypertension (79%), hyperlipidemia (70%), and arthritis (27%). The most common mental health condition was depression (10%). The mean age at cancer diagnosis was (76 ± 7.02) for the low-value care group and (75±6.72) for those without low-value care. The mean fragmentation of care index among individuals who received low-value care was (0.62 ± 0.11) vs. (0.60 ± 0.12) for those without low-value care. The mean number of chronic conditions was higher for the low-value care group (3.67 ± 1.96) vs. (2.78 ± 1.74) for those without low-value care. The rates of low-value care varied by gender for some cancers (NHL and colorectal cancer). Low-value care was higher in women with colorectal cancer (32.7%) vs. (28.8%) compared to men with colorectal cancer. We also observed the highest rates of low-value care in women with NHL (40%) vs. (39%) for men with NHL.

Table 1.

Description of Eligible Adults diagnosed with Breast, Colorectal, Ovarian, Uterine, Prostate Cancers and Non-Hodgkin’s Lymphoma by Low-value care category SEER-Medicare, 2014 – 2015

Characteristic	Low-value care		No Low-value care
	N	%	N	%
Cancer Type
Breast	2764	35.0	7208	38.1
Colorectal	1620	20.5	3367	17.8
Prostate	2510	31.8	6863	36.2
Non-Hodgkin’s
Lymphoma	995	12.6	1495	7.9
Gender
Female	4170	52.9	9776	51.6
Male	3719	47.1	9157	48.4
Age Group
66–69 years	1706	21.6	5256	27.8
70–74 years	2159	27.4	5674	30.0
75–79 years	1746	22.1	3896	20.6
≥ 80 years	2278	28.9	4107	21.7
Race/Ethnicity
White	6653	84.3	15785	83.4
African American	694	8.8	1679	8.9
Others	542	6.9	1469	7.8
Marital Status
unmarried	675	8.6	1616	8.5
married	4118	52.2	10412	55.0
Separated/Widowed/Divor
ced	3096	39.2	6905	36.5
Preventive Influenza Vaccination
Preventive Flu shot	4541	57.6	10093	53.3
No preventive Flu	3348	42.4	8840	46.7
Region
Northeast	1767	22.4	3692	19.5
South	1774	22.5	4770	25.2
North Central	927	11.8	2151	11.4
West	3421	43.4	8320	43.9
Rural/Urban Residence
metro	6893	87.4	15774	83.3
Urban	872	11.1	2789	14.7
Rural	124	1.6	370	2.0
Cancer Stage
Stage 1	2313	29.3	5760	30.4
Stage 2	2976	37.7	7827	41.3
Stage 3	874	11.1	2029	10.7
Stage 4	876	11.1	1711	9.0
Unknown	850	10.8	1606	8.5
Chemotherapy
Chemotherapy	1875	23.8	3942	20.8
No Chemotherapy	6014	76.2	14991	79.2
Number of Chronic Conditions
None	445	5.6	1885	10.0
one-two	1767	22.4	6891	36.4
three or more	5677	72.0	10157	53.6
Income
$8,650–37,401	1453	18.8	3816	20.6
$37,404–49,34	1536	19.9	3845	20.7
$49,360–62,48	1495	19.3	3492	18.8
$62,487–84,23	1525	19.7	3608	19.4
$84,289–250,0	1727	22.3	3798	20.5
	Mean	Standard Deviation	Mean	Standard Deviation
	$49,31
Total Medicare Expenditure	2	$49,669	$32,513	$32,773
Total Out-of-pocket Expenditure	$8,726	$7,214	$6,802	$6,102

Note: Based on 27,067 adults over 66-years, diagnosed with breast, prostate, colorectal, and non-Hodgkin’s lymphoma, living in any of the 18 SEER regions in the United States. All had continuous enrollment for 12 months baseline and 12months follow-up periods. Physical health conditions measured include Asthma, Arthritis, Chronic obstructive pulmonary disease, Chronic Kidney disease, Cardiac arrhythmia, Congestive heart failure, hypertension, hyperlipidemias, Diabetes, hepatitis, dementia.

Out-of-Pocket Expenditure:

Individuals who received one or more low-value procedures had significantly higher mean out-of-pocket expenditure ($8,726±$7,214) vs. ($6,802±$6,102) compared to those who did not have low-value care in the follow-up period. Among all cancer types, individuals who received low-value care had higher out-of-pocket expenditure. The highest mean out-of-pocket expenditure was observed in individuals with NHL and low-value care ($13,554) vs. ($11,698) for those without low-value care. The mean expenditure for breast cancer patients without low-value care was ($6,940±5,780) vs. ($8,412±6,398) for those who had low-value care. The mean expenditure for prostate cancer patients without low-value care was ($5,234±4,238) vs. ($6,878±5278) for those who had low-value care. The mean expenditure also differed by gender for individuals with colorectal cancer. For men with colorectal cancer, the mean out-of-pocket expenditure for those without low-value care was ($8,102±7220) vs. ($9,557±8,299) for those who had low-value care. For women with colorectal cancer, the mean out-of-pocket expenditure for those without low-value care was ($7,064±6,043) vs. ($8,842±7,067) for those who had low-value care. Out-of-pocket expenditure did not differ significantly between men and women with NHL and low-value care ($13,564 vs. $13,540), respectively.

Model Performance

We compared the model performance metrics (Root Mean Square Error [RMSE] and r-squared) between the XGBOOST regression using log-transformed expenditure with XGBOOST Tweedie regression. The log-transformed model performed better (RMSE = 0.52, R² = 0.83) than the Tweedie regression (RMSE = 632, R² 0.71).

Leading predictors of out-of-pocket expenditure

Figure 1 shows the SHAP summary plot for leading predictors of out-of-pocket expenditure. Each point on the plot represents an observation in the data; the x-axis value indicates the effect of the feature on the final prediction. Negative values indicate a negative association with out-of-pocket expenditure, while positive values increase out-of-pocket expenditure. The color gradient represents the value for that feature, with yellow corresponding to lower values of the feature and purple corresponding to higher values. The features are ordered in descending order of importance based on the mean SHAP value. Based on the SHAP summary plot, we observed that cancer, low-value care, number of physical health conditions, and FCI were the top predictors of excess out-of-pocket expenditure. Low-value care increased out-of-pocket expenditure by 5%.

Figure 1: SHAP Summary Plot for Top 15 predictors of Out-of-pocket Expenditure.

Note: Carefrag = Care Fragmentation. age_dx = Age at cancer diagnosis. Median_inc = Median Income.

Mortality3YrMalignantNeoplasm = county level 3-year cancer mortality. MedicareAdvantagePenetration = Medicare Advantage penetration. MecrBenefEligMedcaid = Dual Medicare and Medicaid eligible individuals.

EDVisitsper1kBenef = county level Emergency department visits per 1000 beneficiaries. Phc_nbr = Number of pre-existing chronic conditions. MedcreBenef_Hosp_ReadmissRate = County level Medicare beneficiary hospital readmission rate. FedQualifiedHealthCenters = county level number of Federally qualified health centers. Brst = Breast cancer. Prst = Prostate cancer.

Using SHAP dependence plots (Figure 2), we observed that low-value care increased the out-of-pocket expenditure by $2,000 and up to $4000 for some individuals. This higher out-of-pocket expenditure was also confirmed by the unadjusted and fully adjusted Generalized Linear models (Table 2) and showed that low-value care accounted for $1,470 in excess spending after controlling all other factors. The accumulated local effect plots (Figure 3) show the association of age, physical health conditions, FCI, and median income with out-of-pocket expenditure. The relationship between FCI and expenditure was complex; for values between (0 – 4) and (6 – 8), increasing FCI was associated with higher out-of-pocket expenditure. However, this association was negative for FCI values between 4 and 6. There was an inverse relationship between age and out-of-pocket expenditure, younger adults had higher out-of-pocket expenditure than older adults, and adults over 90 years had on average $2,500 less out-of-pocket expenditure. There was a positive relationship between the number of chronic conditions and out-of-pocket expenditure, and an increasing number of chronic conditions increased expenditure.

Figure 2:

SHAP Dependence Plot for the Association of Low-value care with Out-of-pocket Expenditure

Table 2.

Parameter Estimates from Unadjusted and Adjusted Generalized Linear Models on Out-of-pocket Expenditure among Elderly Medicare Fee-for-service Beneficiaries with Incident Breast, Colorectal, Prostate and Non-Hodgkin’s Lymphoma SEER-Medicare data, 2014–2015 (n=26,590).

	Parameter Estimate (SE)	Change #	Parameter Estimate (SE)	Change #
	Unadjusted Models		Adjusted Model 2
Low-value care ‡
Intercept	8.82 (0.006)	$6,796.72	8.83 (0.03)	$6,857.32
One or more LVCs	0.24 (0.01)	$8,725.87	0.19 (0.01)	$8,327.25
Care fragmentation
Intercept	8.58 (0.02)	$5,323.46	8.83 (0.03)	$6,857.32
Incremental	0.05 (0.00)	$5,614.03	0.03 (0.00)	$7,119.97
Number of Chronic conditions
Intercept	8.8 (0.01)	$6,608.36	9.2 (0.03)	$6,857.32
Incremental	0.03 (0.01)	$6,841.60	0.02 (0.00)	$7,011.56

Note: Based on 26,590 older (age >66 years) Fee-for-Service Medicare beneficiaries, with continuous enrollment in Medicare part A & Part B, diagnosed with Breast, Colorectal, Prostate, and NHL between January 2014 and December 2014. Out-of-pocket expenditures include inpatient, outpatient, durable medical equipment, and home health agency costs.

^‡Composite measure for individuals who received any one of the 20 measured low-value services.

Change # was calculated by the difference between the 1) exponentiation of the model intercept term and 2) the sum of the intercept and the variable parameter estimate.

Compare to omitted category

SE= Standard Error, SEER= Surveillance, Epidemiology and End Results Cancer Registry

Figure 3: Accumulated Local Effect Plot for the association between features and Out-of-pocket Expenditure.

Note: Carefrag = Care Fragmentation. age_dx = Age at cancer diagnosis. Median_inc = Median Income. Phc_nbr = Number of pre-existing chronic conditions.

The features with the most significant interactions with all other features were FCI, age, median income, and the number of chronic conditions were the top interacting features. The feature interactions differed significantly by cancer type. For example, individuals with prostate cancer under 75 years had lower out-of-pocket costs than the rest of the cohort. However, above 75 years, the out-of-pocket costs increased by about $500 to $1,000 above the rest of the cohort. The effect of higher care fragmentation on out-of-pocket expenditure was observed to be significant for FCI values above 0.6 and increases out-of-pocket spending by $500 for every unit increase in FCI. The interaction between low-value care and cancer type varied with the type of cancer (plots not shown). Having low-value care induced a broader variation in expenditure compared to individuals without low-value care for all the cancers.

Discussion

In this first nationwide study of the excess cost burden of low-value care on cancer patients, one in four older adults with cancer received at least one low-value care procedure. The prevalence rates of low-value care differed by cancer type, with the highest rates in patients diagnosed with incident NHL (40%) followed by colorectal cancer (33%), breast cancer (28%), and prostate cancer (27%). The rates of low-value care observed in our study is broadly consistent with previous estimates of low-value care in the Medicare population.^7,8,22,23

Low-value care was associated with higher out-of-pocket expenditure, and we observed that low-value care was driving the excess out-of-pocket expenditure across all cancers after adjusting for other factors. On average, patients who received a low-value procedure experienced between $1,000 and $2,000 higher out-of-pocket expenditure attributable to low-value care. This finding is important because it focuses on the excess cost burden associated with low-value care in cancer patients and its contribution to financial toxicity in these patients.

Multi-pronged interventions targeting patients, physicians, and payers are needed to address this problem of low-value care and the associated economic burden. Patient-focused interventions employing quality metrics and tiered benefit plans²⁴ that lower patient cost-sharing for receiving care from providers and organizations reporting higher quality metrics are needed to help address this problem.²⁵ Previous studies report that patients rarely utilize these reported quality metrics when making decisions about where to receive care,²⁶ primarily because the reports are prepared using technical language that is not easily comprehensible for most patients.^27–29 Provider-focused interventions that inform physicians of the possible out-of-pocket expenditure to patients for specific low-value care have also effectively reduced low-value care.³⁰

This study’s finding that care fragmentation is a top predictor of out-of-pocket expenditure is consistent with previous studies that have established that cancer care in patients with multiple pre-existing chronic conditions is fragmented,³¹ and often occurs in silos.^31–33 Patients with cancer and pre-existing chronic conditions frequently report negative experiences with cancer care due to poor care coordination, medication errors, and financial concerns.^34,35 Fragmented care frequently results in other chronic health needs unintentionally “falling through the cracks” due to poor communication between primary care providers and oncology teams,³⁶ resulting in adverse drug reactions, uncontrolled chronic illness, and excess out-of-pocket expenditure.^37,38 Also, some patients felt providers did not thoroughly review their health records before their appointment, and they were repeatedly asked for the same medical information,³⁴ creating the impression that the provider did not have a clear overview of the patient’s overall medical situation. Effective care coordination that involves the primary care physician from the point of diagnosis through the entire care trajectory is needed to improve outcomes.³⁷ Snyder et al. found that one in five cancer survivors never had a primary care provider visit in the second year following a cancer diagnosis; such gaps in care could result in unmet chronic health needs and an exacerbation of existing chronic conditions.³⁹ In addition, robust clinical guidelines for managing multiple chronic conditions during cancer treatment and cancer survivorship need to be developed. Current guidelines for the management of most chronic conditions exist in silos within the specialty it was developed and do not account for chronic management during cancer survivorship.^40,41

This study’s finding that out-of-pocket spending declined with increasing age is broadly consistent with previous reports that out-of-pocket expenditure initially increases with advancing age then decreases.⁴² However, some other studies have reported increasing out-of-pocket expenditure with increasing age in patients with multiple health conditions.⁴³ Although this study did not investigate the reasons for the observed decline in spending with increasing age, previous studies theorize that the reduction in spending may be associated with decreased access to discretionary health services with increasing age.⁴² We also observed a positive association between the number of chronic conditions and out-of-pocket expenditure in this study. Expenditure remained relatively unchanged between one and four chronic conditions but increased dramatically in individuals with five or more chronic conditions.

Policy Implications

The findings from this study have important clinical and policy implications for various stakeholders in healthcare. Specifically, the Centers for Medicare and Medicaid Services (CMS), with over 65 million enrollees, has issued a roadmap for states to accelerate the adoption of value-based care, with the Department of Health and Human Services (HHS) setting a goal of converting 50% of traditional Medicare payment systems to alternative payment models tied to value-based care by 2022.⁴⁴ In the 2021 Notice of Benefit and Payment Parameters rule, the department of Health and Human Services encouraged the use of value-based insurance plans with consumer cost sharing levels that promote the use of high value services while discouraging low-value care.⁴⁵

Future Directions

Although the present work focused on the prevalence and economic burden of low-value care on patients, a more holistic approach that examines the humanistic, clinical, and social burden is needed. Future studies using EHR data and machine learning approaches can further characterize the processes of low- and high-value care delivery by incorporating clinical factors. It has to be noted that using linked registry and claims data; this study did not control for observable and unobservable selection bias in the evaluation of low- and high- value care, the Covid-19 pandemic may have created a natural experiment to study the “real effects” of low- and high-value care because, in many health systems, elective procedures were halted instantaneously creating randomization of patients into low-value and no-treatment groups.

Future studies can employ “deep learning” AI methods to further elucidate the predictive relationships between personal characteristics and low- and high-value care to personalize care and intervention efforts.

Strengths and limitations

This is the first population-based study to quantify the direct economic impact of non-cancer low-value care in older adults diagnosed with cancer and fills a critical gap in the literature. This study uses real-world data from a patient registry that is nationally representative of adults over 65 years with cancer. We employed advanced machine learning algorithms to identify the leading predictors of the excess cost burden and used model agnostic explanation tools to explain the machine learning predictions.

This study has several limitations; like all direct measures of low-value care, our estimates may be limited because of the quality of available data. Furthermore, there is a potential for misclassifying some instances of low value-care in instances where care was appropriate and high value because of the lack of more detailed clinical information often required to make treatment decisions. We were only able to measure a limited number of the over 150 low-value care services in the choosing wisely recommendations because some of the recommendations on the list cannot be measured using the available data, so we may not have captured the most critical low-value services for health system efficiency and patient outcomes.

Conclusion

One in four older adults received low-value care in the follow-up period. Across all cancer populations, individuals who received low-value care had significantly higher out-of-pocket expenditure. Excess out-of-pocket expenditure was driven by low-value care, fragmentation of care, and a higher number of pre-existing chronic conditions. Policy initiatives that target low-value care and fragmented care may reduce the financial burden on the payers and patients.

Highlights.

• This study focused on the excess cost burden associated with low-value care in cancer patients and its contribution to financial toxicity in these patients.

• One in four older adults with incident cancer received one or more low-value procedures in the follow-up period.

• Low-value care was associated with higher out-of-pocket expenditure and was driving the excess out-of-pocket expenditure across all cancers after adjusting for other factors.