This is a retrospective cohort study using Louisiana Medicare Fee-for-Service claims linked to beneficiary files and analyzed using an inverse probability treatment weighted (IPTW) controlled interrupted time series (CITSA) design. This manuscript was prepared following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines [43].

We used data from a patient-level longitudinal panel from January 1, 2016, to December 31, 2020, which was constructed from Louisiana Medicare Fee-for-Service claims and Beneficiary files. This data is collected at the time of service and was purchased from the Research Assistance Data Center for secondary analyses. We included patients who had a diagnosis of diabetes or hypertension based on the CMS Chronic Conditions files. Patients were excluded if they did not have a diabetes or hypertension diagnosis before 2016, were not enrolled in Medicare before 2016, had only gestational diabetes or secondary diabetes, or had end-stage renal disease. These exclusions were found using the ICD codes.

Patients were included in the treatment group if they had at least one occurrence of a common procedure terminology (CPT) code for RPM or continuous glucose monitoring (99453, 99454, 99457, 99458, 95250, 95251, 99091, E2100, E0607, E2102, E2103, E2101). This information was found in the inpatient, ER, or outpatient claims files. We defined the treatment month as the first month with this CPT code. For the control group, pseudo-treatment dates were randomly assigned to reflect the distribution of treatment initiation in the intervention group. Specifically, in Excel, each month and year combination was coded numerically, and controls were assigned a treatment month and year in proportion to the observed distribution of initiation dates in the treatment group. This ensured alignment by time of year rather than only by year, reducing bias from differential calendar-time exposure. After the assignment, we conducted balance checks and confirmed that the month-by-month distribution of treatment dates did not significantly differ between the control and treatment groups.

The number of outpatient, ER, or inpatient visits per month. We defined these outcomes using published guidance from the Research Data Assistance Center using revenue center codes, claim dates, and places of service codes in the Medicare Fee-for-Service files (Carrier claims, Carrier Line, Base Claims, and Revenue Center). The disease-specific outcomes included the number of outpatient, ER, or inpatient visits per month where the primary ICD code indicated that the visit was for diabetes or hypertension. If a beneficiary did not have an encounter in a certain month, they were coded as having 0 visits. Visits with RPM as the primary CPT code were not included. All outcomes were detrended to remove underlying time trends to focus on deviations that may be associated with RPM.

We used a CITSA to determine the impact of RPM. This approach leverages pre- and posttreatment data from a treatment and control group to estimate changes in the level and slope of the outcome variables before and after treatment initiation. This method assesses the impact of an intervention such as RPM by comparing a treatment to a control group over time. CITSA was chosen for the analysis because of the linearity of the data over time (Multimedia Appendix 1) and for its ability to account for pre-existing trends and provide a counterfactual estimate. The CITSA model may be written as:

The baseline level is given by the intercept β₀. β₁ represents the underlying time trend for the non-RPM group before the intervention. The indicator variable RPM specifies group membership, with β₂ capturing any baseline differences between the intervention and control groups. The term β3 allows for differences in preintervention trends between RPM and non-RPM groups. The coefficient β₄ represents the immediate postintervention effect. The 3-way interaction term β₅ captures any differential trend change attributable to the intervention over time. There was no specific postperiod variable included in the model, as the control group had a randomly assigned treatment date and, as such, had no expected postperiod outcome change.

To account for bias and confounding, we adjusted for covariates that may influence the outcome, including comorbidities (diabetes, hypertension, COPD, chronic kidney disease, hyperlipidemia), Medicare status, sex, race (White, Black, Hispanic, Asian, Other), rurality (defined as metropolitan, micropolitan, rural using FIPS codes and 2020 core-based statistical area data), and age. These variables were created using the Medicare Beneficiary Files (chronic condition and ABCD files), which aggregate data at the annual level.

All analyses accounted for time- and beneficiary-level fixed effects and were weighted by the IPTW, and standard errors were clustered at the primary doctor level to account for serial correlations over time. To construct the weights, we modeled the probability of RPM participation using baseline demographic and clinical characteristics from the month before treatment and the average preperiod outcomes. Each beneficiary was assigned a weight equal to the inverse of their predicted probability. This data was obtained from the beneficiary files and the previously created outcome variables. The primary doctor was defined as the physician whom a beneficiary saw the most during the study period to account for correlated practice patterns. In the base case, the main model used all available data in the 60 months regardless of pre- and postlength of time.

We used an inverse-probability–weighted 2-way fixed-effects model to control for unobserved, time-invariant differences across beneficiaries and shared temporal shocks. This design isolates within-beneficiary changes before and after RPM initiation, consistent with the interrupted time series framework. It was preferred to multilevel models because exposure timing varies across beneficiaries, and fixed effects more clearly estimate pre/post level and trend differences [44].

Because Medicare claims are used for reimbursement, demographic, enrollment, and utilization fields are complete for almost all beneficiaries. Diagnosis and procedure variables used to construct comorbidities are also not subject to item-level analysis. Multiple imputation was not performed, as missingness was minimal and largely structural and less than 15% for all key variables.

All confidence intervals were calculated at the 95% confidence level. We tested the robustness of the results by clustering standard errors by zip code and changing the necessary length of pre- and posttreatment data to 6 months pre and 6 months post, and 12 months pre and 12 months post. Additionally, we assess seasonality of treatment by running the analysis with a variable for calendar month (January, February, etc) and another analysis with a control variable for calendar month of treatment. Additionally, we ran an analysis excluding COVID years to ensure we were not picking up COVID confounding and to account for potential exposure classification issues due to new CPT codes being introduced.

All analyses were conducted using Stata (version 18.0; IBM Corp).

This study was approved by Tulane University’s Institutional Review Board (2023-1044-SPHTM) through an expedited review, as it only uses secondary data. The data use agreement and institutional review board for this study allow the use of secondary data analysis without additional consent from beneficiaries. All study data is deidentified to protect beneficiaries, and there was no way to reidentify. As we did not work directly with beneficiaries, no compensation was given, and there is no identification of individuals in this manuscript.

This study aimed to assess the impact of RPM on HCRU on Louisiana Medicare beneficiaries with diabetes or hypertension from 2016 to 2020. By combining a rigorous longitudinal design with a focus on an older, high-burden Medicare population, this study provides novel insights into how RPM adoption affects health care utilization in real-world settings. Overall, in beneficiaries with diabetes or hypertension, RPM was associated with a statistically significant increase in outpatient, ER, and inpatient visits immediately after initiation, followed by a significant downward trend over time. These findings, supported by narrow confidence intervals, suggest that the estimates are precise and due to RPM, not random variation.

Existing evidence on RPM’s impact on utilization is mixed. Some studies report decreases in HCRU [35,45]. Others have shown that it may not change utilization [17]. Notably, studies in a Medicare population also demonstrated an increase in utilization; for example, one study on Medicare hypertensive patients reported an increase in cardiovascular outpatient visits [38]. Differences across studies may reflect variation in patient populations, RPM modalities, and social determinants of health. For example, our Medicare population may have lower digital or health literacy.

Additionally, the average beneficiary in our cohort was on RPM for 7 months, and results were not robust to a short time period, which is consistent with literature that found that clinical outcomes, such as decreased blood glucose, which may impact utilization, are only found in longer time periods [46]. Thus, long-term users may be driving the observed effects, underscoring the importance of sustained participation for realizing RPM’s full impact.

The dynamic trend of early increases followed by subsequent declines in utilization aligns with the literature, suggesting that individuals initially engage more intensively with new tools before stabilizing into routine patterns [47,48]. In the initial phase of RPM adoption, both patients and providers may be adjusting to new data and workflows. Patients may respond to frequent health alerts or health data with heightened awareness, prompting additional interactions with the health system. Providers may proactively schedule follow-ups to verify readings, resolve technical issues, or re-evaluate treatment plans due to the additional data.

The subsequent decrease might indicate that as patients become more accustomed to the data, they take a more measured approach to seeking care. It remains unclear whether these patterns represent appropriate, preventive care or unnecessary and wasteful utilization. The increase could stem from useful data and engagement that identify unmet needs and ensure patients are getting necessary care, or it could represent low-value encounters due to the large influx of data or cautious clinical behavior. Without information on clinical appropriateness, clinical outcomes, and more information on the visits, it is not possible to determine whether the short-term increase benefits patients and the health system.

This observed pattern likely reflects a complex interaction of patient and provider behaviors shaped by differences in digital literacy, clinical workflows, and familiarity with RPM. For instance, improved patient self-management is one of the key benefits of RPM and can lead to better outcomes, but it is dependent on patients’ ability to understand health data [49,50]. Both health and digital literacy are associated with age and education [50]. Given that this study focused primarily on older adults in a state with relatively low average education levels, some patients may have lacked the digital or health literacy required to meaningfully engage with RPM. This could limit the intervention’s potential benefits or even contribute to uncertainty-driven care-seeking rather than reducing visits.

Additionally, our study focused on early RPM adopters and physicians who were still learning effective RPM implementation methods. Clinicians were still learning how to interpret continuous data, integrate RPM into workflows, and communicate with patients. A systematic review of factors of RPM success highlights several physician-level factors for effectiveness, including targeting populations at high risk, providing responsive and timely care, and ensuring collaborative and coordinated care [45]. If providers are learning how to provide timely and coordinated care via new technologies, RPM may not have its intended benefit. Together, these patient- and provider-level factors suggest that the short-term rise in utilization may reflect a broader adaptation period as both groups learned to navigate new technologies and patterns of care.

Moreover, this study focuses specifically on Louisiana Medicare. The state has a relatively high chronic disease burden, an older average population, and low levels of health and digital literacy [51,52]. These are all factors that can slow early adoption and increase initial contact with providers as patients and clinicians adapt to new tools. At the same time, Louisiana’s high poverty rate and large rural population may limit broadband access [53], contributing to declining utilization after initial implementation phases stabilize. Finally, reimbursement for RPM under other payers (Medicaid and commercial programs) was only gradually expanded in Louisiana after 2019 as well, constraining early uptake and shaping provider behavior during the study period. These contextual features suggest that both the learning period and structural factors unique to Louisiana likely influenced the temporal dynamics observed in this study.

From a policy and clinical perspective, these findings have dual and uncertain implications. While increased short-term utilization may raise costs, if the increased visits reflect timely, preventive, or necessary care, the investment could yield downstream savings through decreased complications and their associated treatments. Conversely, if the increased encounters are duplicative or not clinically necessary, they could add costs. Policymakers and payers should therefore interpret early utilization spikes with caution. To prove that these visits should be encouraged, future research should further assess costs and clinical endpoints, which can clarify whether observed utilization patterns translate into value-based outcomes.

In the meantime, these results do offer some recommendations on how to maximize the benefits of RPM and potential challenges that need to be mitigated. First, providers should prioritize patient education initiatives; ensure that patients, especially older adults in Medicare, receive training on how to collect, interpret, and respond to RPM data. This finding is supported by a recent systematic review showing that RPM was more effective when patients had education and support in interpreting data [22]. Education programs can help enhance patient engagement and self-management while preventing the misuse of health care resources. Additionally, health care providers need structured support, including training on integrating RPM into care delivery and systems in place, to ensure that they have the resources to analyze RPM data.

As this study is observational in nature, the ability to assess causality is limited due to the potential for confounding and the lack of randomization. There is an inherent bias in how patients receive RPM. While the methodology attempts to overcome this, it may not have been able to completely eliminate biases in patient selection, which could bias the results away from the null because RPM patients are already engaging with the health system. Similarly, we handled confounding through IPTW and using controls, but some confounding may still be present. Additionally, the findings have limited generalizability as the population with diabetes and hypertension in Louisiana Medicare may differ from other populations. Therefore, these results may not be applicable when assessing or developing policy for other settings, including younger patients, other insurance types, or in other states.

Similarly, this study focused only on RPM through 2020, which was before the large increase in RPM and telehealth utilization due to COVID-19, and the study results may not be the same as those reported today, as some of the barriers mentioned here are less of a concern, resulting in results biased towards the null.

Moreover, this study did not assess the duration of RPM utilization or the number of times a physician checked the RPM data. Claims data do not allow assessment of patient engagement with RPM, and this study only looks at RPM prescription, not true utilization, as an outcome, although actual use mitigates the impact. Therefore, this might bias results towards the null if this group is not engaging with the technology or bias upwards if this population is engaged with RPM technology.

Similarly, the RPM CPT code does not indicate the type of device a patient is using, and as a result, these findings cannot be applied to certain RPM types, such as continuous glucose monitoring, and represent the larger impact of RPM as a treatment option. Finally, a significant limitation of this study is that it did not assess the appropriateness of utilization or the impact on clinical outcomes. If the increase in utilization stems from patients receiving timely care or appropriate care, health outcomes may improve.

This study is innovative in its use of comprehensive real-world evidence from statewide Medicare claims to evaluate RPM among older adults with diabetes or hypertension, a population underrepresented in prior studies. The analysis distinguishes short-term onboarding effects from longer-term stabilization in health care use, expanding on potential reasons for inconsistent findings in earlier research. Conducted during a period of expanding RPM reimbursement, these results provide timely insights for clinicians and payers on targeting, education, and workflow integration to optimize RPM value and reduce low-value encounters.

This study assessed the association between RPM and HCRU among Louisiana Medicare beneficiaries with diabetes or hypertension. We found that RPM was associated with a modest but statistically significant increase in health care utilization, highlighting important questions about how RPM is implemented and integrated into routine care. This study contributes to the existing literature by examining RPM at scale in a real-world Medicare population using administrative claims data, rather than focusing on short-term clinical outcomes or small pilot programs. Unlike prior studies that primarily evaluate changes in biometric measures, our findings underscore the need to consider utilization patterns as an early indicator of how RPM influences care delivery. Although the current analysis could not assess the clinical appropriateness or downstream value of these encounters, the results suggest that initial increases in utilization may reflect enhanced monitoring, care coordination, or unmet need. Future work should link claims with clinical and cost data, examine patient engagement and digital literacy, and identify implementation strategies that ensure RPM supports high-value, equitable care delivery.