Morbidity, mortality, facility placement and cost of deep brain stimulation for Parkinson’s disease

Abstract

Deep brain stimulation (DBS) is an established therapy for advanced Parkinson’s disease (PD), but its long-term social and clinical impact has not been well characterized in large cohorts. We conducted a retrospective cohort study using a nationwide hospital database across the United States. Among 79,845 adults hospitalized with PD in 2018 and followed through 2023, 482 underwent DBS implantation. Overlap propensity score weighting was used to balance baseline characteristics. Weighted logistic and gamma regression models estimated adjusted odds ratios (aORs) and cost ratios at 1, 3, and 5 years after discharge. DBS was associated with significantly lower risks of institutionalization at 1, 3, and 5 years (aORs 0.25–0.27) and lower mortality at 1 and 5 years (aORs 0.09 and 0.27). Total healthcare expenditures were also lower in the DBS group (cost ratios 0.68–0.73). Readmission rates were similar between groups but tended to decline over time among DBS recipients. Sensitivity analyses excluding patients with dementia and restricting to non-emergency admissions yielded consistent results. Although DBS implantations were concentrated in large, urban teaching hospitals, outcomes were comparable between teaching and non-teaching institutions. These findings suggest sustained clinical and economic benefits of DBS and highlight the importance of improving equitable access.

Introduction

Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by motor and non-motor complications that frequently lead to hospitalization and long-term care especially as the disease advances¹. Deep brain stimulation (DBS) is an established surgical therapy for select patients with medication-refractory PD and has been shown to provide sustained improvement in both motor symptoms and quality of life^2–4. Despite its well-documented clinical benefits, DBS is an invasive intervention that can be associated with perioperative and hardware-related complications and these have been associated with unplanned hospital readmissions^5–8. Furthermore, long-term outcomes such as institutionalization, mortality, and healthcare expenditures among DBS recipients have been variably reported across studies, with inconsistencies reflecting differences in patient selection, comorbidity burden, and individual differences in healthcare systems^9,10. To date, comprehensive evaluations of these outcomes using large-scale, nationally representative data remain scarce. Using the PINC-AI™ Healthcare Database, a nationwide claims-based dataset in the United States (U.S.), this study examined the long-term impact of DBS on hospitalization, institutionalization, mortality, and medical costs in PD. By applying overlap weighting based on propensity scores, we sought to clarify the real-world clinical and economic implications of DBS over 1, 3, and 5 years follow-up periods.

Results

Patient characteristics

Among hospitalized patients with PD who met the inclusion criteria, 482 (0.6%) underwent DBS during the index admission (DBS group), while 79,363 (99.4%) did not (non-DBS group). Baseline characteristics of patients with PD according to DBS status are shown in Table 1. Before weighting, patients in the DBS group were substantially younger (mean ± SD = 66.6 ± 7.6 years vs. 76.9 ± 8.6 years; SMD = 1.28) and more frequently male (67.0% vs. 59.0%). DBS recipients had lower rates of dementia (2.3% vs. 33.6%; SMD = 0.89) and markedly lower comorbidity burden (CCI 0.24 ± 0.66 vs. 1.96 ± 1.90; SMD = 1.21). Most DBS implantations were performed in large, urban, teaching hospitals. Among DBS group, 98.1% were admitted to urban facilities (vs 88.3%), and 86.9% were treated at teaching hospitals (vs. 46.9%). Admissions were primarily elective from home or outpatient clinics, and DBS was most frequently performed in hospitals with ≥ 500 beds (66.4% vs 31.0%). Geographically, DBS utilization was highest in the South (57.1%), followed by the Northeast (19.7%), Midwest (17.6%), and West (5.6%). Patients receiving DBS were more likely to have private insurance (30.1%) and less likely to be covered exclusively by public programs such as Medicare or Medicaid (68.7%) when compared with non-DBS patients (6.3% and 92.9%, respectively).

Table 1.

Baseline characteristics of PD patients by DBS status

Variable	Level	No DBS (n = 79,363)		DBS (n = 482)		SMD
Age, mean (SD)		76.9 (8.6)		66.6 (7.6)		1.276
Gender	Female	32,539	41.00%	159	33.00%	0.167
	Male	46,824	59.00%	323	67.00%
Race group	Asian	1508	1.90%	13	2.70%	0.251
	Black	5000	6.30%	8	1.70%
	Other	4444	5.60%	36	7.50%
	White	67,300	84.80%	417	86.50%
Marital status	Alone	41,745	52.60%	216	44.80%	0.156
	With others	37,618	47.40%	266	55.20%
Admission origin	Clinic	6270	7.90%	180	37.30%	0.863
	Home	61,506	77.50%	293	60.80%
	Other	1190	1.50%	3	0.60%
	Transfer	10,397	13.10%	6	1.20%
CCI score, mean (SD)		1.96 (1.90)		0.24 (0.66)		1.213
Dementia	No	52,697	66.40%	471	97.70%	0.893
	Yes	26,666	33.60%	11	2.30%
Urban / Rural	Rural	9285	11.70%	9	1.90%	0.398
	Urban	70,078	88.30%	473	98.10%
Teaching hospital	No	42,142	53.10%	63	13.10%	0.939
	Yes	37,221	46.90%	419	86.90%
Bed size	0–99	5000	6.30%	0	0%	0.896
	100–199	12,301	15.50%	36	7.50%
	200–299	14,047	17.70%	15	3.10%
	300–399	13,809	17.40%	56	11.60%
	400–499	9524	12.00%	55	11.40%
	>500	24,603	31.00%	320	66.40%
Region	Midwest	18,888	23.80%	85	17.60%	0.326
	Northeast	15,635	19.70%	95	19.70%
	South	34,999	44.10%	275	57.10%
	West	9841	12.40%	27	5.60%
Payor group	Other	714	0.90%	6	1.20%	0.653
	Private	5000	6.30%	145	30.10%
	Public	73,728	92.90%	331	68.70%

Values are presented as mean (SD) or n (%). Standardized mean differences (SMD) > 0.1 indicate meaningful imbalance between groups.

Covariate balance after overlap weighting

Before weighting, notable differences were observed across multiple baseline characteristics. After applying overlap weighting, excellent covariate balance was achieved for all measured covariates (standardized mean difference [SMD] <0.01 for every variable), confirming effective adjustment. The effective sample sizes after weighting were 474 (DBS group) and 5321 (non-DBS group), indicating stable weighting without loss of precision. The Love plot (Fig. 1) demonstrated a marked improvement in covariate balance, with all post-weighting SMDs <0.1.

Fig. 1. Covariate balance before and after overlap weighting.

Standardized mean differences (SMDs) for all baseline covariates are shown before and after overlap weighting. Overlap weighting achieved excellent balance across all variables, with all adjusted SMDs falling within the ±0.1 threshold. DBS deep brain stimulation, CCI Charlson Comorbidity Index.

Readmission

Overlap-weighted logistic regression revealed no significant differences in readmission between groups (Table 2). At 1 year, the estimated readmission rate was 37.5% (95% CI, 33.1–41.8%) for DBS group vs. 34.3% (33.1–35.5%) for non-DBS group (p = 0.17). At 3 years, 42.8% (38.3–47.2%) vs. 43.6% (42.3–44.9%) (p = 0.72), and at 5 years, 45.2% (40.7–49.7%) vs 47.6% (46.2–48.9%) (p = 0.32). Although the differences did not reach statistical significance, the DBS group showed a tendency toward progressively lower readmission rates over time. The rate at 1 year was slightly higher in the DBS group, but the gap gradually widened at 3 and 5 years, suggesting a potential long-term reduction in readmissions compared with the non-DBS group (Fig. 2A).

Table 2.

Unadjusted outcomes by DBS status at 1, 3, and 5 years

Outcome	No DBS (n = 79,363)	DBS (n = 482)
Readmission within 1 year	35,981 (45.3%)	177 (36.7%)
Readmission within 3 years	41,795 (52.7%)	202 (41.9%)
Readmission within 5 years	43,403 (54.7%)	214 (44.4%)
Institutionalization within 1 year	20,545 (25.9%)	19 (3.9%)
Institutionalization within 3 years	24,588 (31.0%)	25 (5.2%)
Institutionalization within 5 years	25,938 (32.7%)	31 (6.4%)
Death within 1 year	1819 (2.3%)	0 (0.0%)
Death within 3 years	2659 (3.4%)	3 (0.6%)
Death within 5 years	2954 (3.7%)	3 (0.6%)
Total cost within 1 year (median [IQR], USD)	$20,680 [7959–55,310]	$21408 [15833–37609]
Total cost within 3 years (median [IQR], USD)	$25,533 [9558–69,412]	$22469 [15739–48583]
Total cost within 5 years (median [IQR], USD)	$27,439 [10234–74,989]	$26001 [16402–52438]

Values are presented as n (%) for binary outcomes and as median [IQR] in U.S. dollars (USD) for cost outcomes. Readmission, institutionalization, and death were evaluated at 1, 3, and 5 years after discharge. Total cost represents the cumulative all-cause medical expenditure incurred within each time window. Shapiro–Wilk tests indicated non-normal distribution for all cost variables (p < 0.001); therefore, median [IQR] values are reported instead of mean ± SD.

Fig. 2. Long-term outcomes after DBS versus no DBS in PD.

Panels show overlap propensity score weighted estimates with 95% confidence intervals at 1, 3, and 5 years after discharge. A Hospital readmission. B Institutionalization (admission to a skilled nursing facility, intermediate care facility, or long-term care institution). C All-cause mortality identified from subsequent inpatient, outpatient, or facility claims. D Total medical costs (inpatient, outpatient, and long-term care expenditures) during each follow-up interval; analyses were restricted at each timepoint to patients with nonzero observed costs. Between-group comparisons were performed using overlap-weighted logistic regression for binary outcomes and overlap-weighted generalized linear models with a log link and gamma distribution for cost outcomes. DBS deep brain stimulation.

Institutionalization

After overlap weighting, Firth penalized logistic regression demonstrated that DBS was independently associated with a significantly lower risk of subsequent institutionalization. The aORs for institutionalization were 0.26 (95% CI, 0.15–0.44) at 1 year, 0.25 (0.15–0.40) at 3 years, and 0.27 (0.18–0.42) at 5 years (all p < 0.001). The DBS group showed consistently lower institutionalization rates across all follow-up periods (Fig. 2B).

Mortality after discharge

Weighted penalized logistic regression indicated that DBS was associated with a persistently lower risk of death across follow-up intervals. The adjusted ORs (DBS vs non-DBS) were 0.09 (95% CI 0.0007–0.81; p = 0.028) at 1 year, 0.34 (0.08–1.12; p = 0.077) at 3 years, and 0.27 (0.07–0.85; p = 0.023). These findings suggest a durable long-term survival advantage in the DBS group (Fig. 2C).

Weighted medical costs

After applying overlap weighting and restricting each timepoint analysis to patients with nonzero observed costs, weighted generalized linear models with a log link and gamma distribution were used to estimate total healthcare expenditures. Across all follow-up periods, total medical costs were significantly lower in the DBS group compared with the non-DBS group (Fig. 2D). At 1 year, weighted mean costs were $36,644 (95% CI $30,747–$42,541) for the DBS group versus $50,543 (95% CI $45,037–$56,050) for the non-DBS group (cost ratio 0.73, 95% CI 0.60–0.88; p = 0.0012). At 3 years, costs were $45,310 ($37,057–$53,563) vs. $66,583 ($60,540–$72,626) (cost ratio 0.68, 0.56–0.83; p < 0.001), and at 5 years, $50,555 ($40,224–$60,886) vs. $74,469 ($68,123–$80,815) (cost ratio 0.68, 0.54–0.85; p = 0.0006).

Additional analyses to assess cohort comparability

To better characterize the non-DBS cohort and address potential differences in cohort comparability, we conducted additional analyses. Using unweighted data, we examined the principal diagnoses associated with index hospitalizations in the non-DBS group and found that the most common diagnoses included sepsis, pneumonia, urinary tract infection, acute kidney failure, and hip fracture. At a broader level, these diagnoses were predominantly classified as infections, respiratory diseases, urinary conditions, and musculoskeletal injuries, indicating that many hospitalizations were driven by acute medical conditions rather than PD itself.

We then performed sensitivity analyses to assess the robustness of our findings. Excluding patients with dementia yielded results largely consistent with the primary analysis. The risk of institutionalization remained significantly lower in the DBS group, with an odds ratio of 0.25 at 1 year (p < 0.001), and risk ratios of 0.28 (p < 0.001) and 0.31 (p < 0.001) at 3 and 5 years, respectively. Mortality was also significantly lower at 3 and 5 years (odds ratio 0.26, p = 0.022; odds ratio 0.22, p = 0.009), and healthcare costs were consistently reduced across all time points (cost ratios 0.46, 0.46, and 0.44 at 1, 3, and 5 years, respectively; all p < 0.001). In contrast, readmission rates did not differ significantly.

Further restricting the non-DBS cohort to non-emergency admissions to reduce the influence of acute illness yielded similar findings. Institutionalization remained significantly lower in the DBS group, with an odds ratio of 0.38 at 1 year (p < 0.001), and risk ratios of 0.38 (p < 0.001) and 0.41 (p < 0.001) at 3 and 5 years, respectively. Healthcare costs were also significantly lower (cost ratios 0.53, 0.55, and 0.55; all p < 0.001). Readmission was higher at 1 year (odds ratio 1.31, p = 0.009) but not at later time points, and mortality showed a decreasing trend without statistical significance. Overall, the direction and magnitude of associations for institutionalization and healthcare costs remained consistent across all sensitivity analyses.

Subgroup analysis by hospital teaching status

Among patients with an index hospitalization (n = 482), 419 (86.9%) were admitted to teaching hospitals and 63 (13.1%) to non-teaching hospitals. Patients treated in teaching hospitals were slightly younger and predominantly admitted to large, urban facilities, particularly in the South and Northeast regions. Comorbidity burden, marital status, and insurance distribution were generally similar between the two groups. Detailed baseline characteristics are provided in Supplementary Table 1. No material differences were observed in post-discharge outcomes between teaching and non-teaching hospitals. The odds of hospital readmission at 1, 3, and 5 years after discharge were modestly higher in teaching hospitals (aOR ranging from 1.29 to 1.58), but all 95% confidence intervals included unity. Total medical costs at 1, 3, and 5 years were comparable between the two groups (cost ratios ranging from 0.71 to 0.86). Institutionalization and mortality were not analyzed because of the limited number of events.

DBS utilization and sociodemographic disparities

This section presents an analysis examining nationwide and sociodemographic patterns of DBS utilization among all inpatients with PD who underwent DBS during the entire period covered by the database (n = 3589). The mean age was 67.0 ± 7.5 years, and 2491 (69.4%) were male. Most were White (85.3%) and resided in urban areas (98.9%). Procedures were primarily performed at teaching hospitals (89.6%) with large bed capacity (≥500 beds in 66.6%). Public insurance covered 72.6% and private insurance 25.8% of cases. Regionally, DBS utilization was highest in the South (42.1%), followed by the Northeast (24.5%), West (17.1%), and Midwest (16.2%) (Supplementary Table 2).

Factors associated with DBS utilization

We fit a multivariable logistic regression across all hospitalized PD patients in the PINC-AI™ Healthcare Database, with DBS receipt (yes/no) as the dependent variable. Younger age, male sex, treatment at teaching hospitals, urban residence, and lower comorbidity burden were independently associated with higher odds of undergoing DBS. Compared with the Midwest, patients in the Northeast (OR = 2.61), South (1.65), and West (3.56) were significantly more likely to receive DBS.

Private (OR = 2.01) and public insurance (1.49) were both associated with greater DBS utilization compared with other insurance types (Table 3).

Table 3.

Multivariable logistic regression identifying factors associated with DBS utilization

Variables	OR (95% CI)	p value
Age (per year)	0.92 (0.92–0.93)	<0.01
Black (vs Asian)	0.44 (0.32–0.59)	<0.01
CCI score (per point)	0.58 (0.55–0.60)	<0.01
Insurance: Private (vs Other)	2.01 (1.55–2.64)	<0.01
Insurance: Public (vs Other)	1.49 (1.16–1.96)	<0.01
Intercept	0.01 (0.01–0.02)	<0.01
Male (vs Female)	1.40 (1.30–1.51)	<0.01
Other (vs Asian)	1.48 (1.17–1.88)	<0.01
Region: Northeast (vs Midwest)	2.61 (2.35–2.91)	<0.01
Region: South (vs Midwest)	1.65 (1.50–1.82)	<0.01
Region: West (vs Midwest)	3.56 (3.16–4.01)	<0.01
Teaching hospital (vs non-teaching)	7.60 (6.82–8.50)	<0.01
Unknown (vs Asian)	1.21 (0.90–1.61)	0.203
Urban (vs Rural)	5.33 (3.96–7.38)	<0.01
White (vs Asian)	1.36 (1.11–1.67)	<0.01

Odds ratios (OR) and 95% confidence intervals (CI) are presented.

Variables were selected a priori based on clinical and demographic relevance.

Discussion

In this large, national cohort of hospitalized patients with PD, we found that individuals who underwent DBS had different long-term outcomes compared with those without DBS. Before adjustment, DBS recipients were younger, had fewer comorbidities, and were more likely to be treated at large, urban, teaching hospitals. After adjustment with overlap weighting, DBS was associated with lower rates of institutionalization and mortality after discharge. Hospital readmission rates were comparable between the two groups. Total healthcare expenditures were consistently lower in the DBS group across all follow-up periods. At the same time, DBS utilization was disproportionately higher among patients who were younger, male, White, privately insured, and treated at academic medical centers, suggesting regional and socioeconomic disparities in access.

While DBS is an established therapy with well-documented motor and economic benefits, prior studies have largely focused on short-term outcomes, motor symptoms, or cost-effectiveness, with limited evaluation of long-term, population-level social endpoints. For example, nationwide analyses have primarily reported short-term readmission outcome¹¹, while long-term outcomes such as mortality and institutionalization have been examined in smaller or regional cohorts with inconsistent findings⁹. Using a nationwide claims database, our real-world analysis extends these prior findings by demonstrating the long-term impact of DBS across multiple clinically meaningful outcomes, including institutionalization, mortality, and healthcare utilization over a 5-year period. It is associated with a reduced risk of institutionalization and mortality, along with sustained containment of healthcare expenditures. These findings are both mechanistically and clinically consistent. Stabilization of motor fluctuations and reduction in dopaminergic medication burden may possibly mitigate downstream complications such as falls, aspiration pneumonia, contractures, and delirium, which contribute to loss of independence, while preservation of physical activity and self-care capacity may possibly delay the need for long-term caregiving^12–14. Contemporary programming strategies and device advances further limit cumulative treatment burden, allowing the benefits to translate into long-horizon resource savings^15–17.

The impact of DBS on survival remains uncertain in prior literature^18–22. Some observational studies have reported a survival benefit, whereas others have found no significant difference^4,9. Reviews of long-term outcomes suggest that any potential effect of DBS on mortality or disability milestones, including institutionalization, remains difficult to establish due to methodological limitations²³. Prior studies have shown that PD is associated with increased mortality compared with the general population, with a standardized mortality ratio of approximately 1.5 and a median survival of about 11.8 years in real-world cohorts^24,25. Reported 5-year mortality after DBS ranges from 10 to 20%, and some studies have suggested that DBS may be associated with reduced mortality in selected patients^9,10. Although the direction of association was consistent with prior reports, the 5-year post-discharge mortality in our study was markedly lower at 0.6% in the DBS group and 3.7% in the non-DBS group. This discrepancy likely reflects reliance on death records captured within subsequent inpatient, outpatient, or facility claims, which may fail to capture deaths occurring outside the database or after loss of follow-up. In aggregate, our findings extend prior observations to population-scale associations involving functional outcomes and healthcare utilization, while supporting the need for cautious interpretation given the limitations of claims-based data.

Prior studies have shown that DBS can be cost-effective, primarily based on modeled analyses incorporating quality-adjusted life years and long-term care reductions²⁶. Real-world analyses further suggest that device-related factors, such as rechargeable systems, may reduce long-term costs by avoiding repeated surgical replacements and associated complications²⁷. In this context, our study extends prior findings by demonstrating that DBS was associated with lower long-term healthcare expenditures in a nationwide real-world cohort, suggesting that economic benefits observed in modeled analyses may translate into routine clinical practice. As our analysis was based on direct healthcare expenditures without incorporating quality-adjusted life years, it provides a complementary perspective to prior cost-effectiveness studies. However, as these cost estimates are derived from the U.S. healthcare system, they may not be directly generalizable to other settings.

The persistence of readmission rates likely reflects a mixture of planned, clinically appropriate encounters, generator replacements and programming visits, together with unplanned admissions, a distinction not separable in claims, and one that can mask reductions in disease-driven hospitalizations^28,29. Prior studies have largely evaluated readmission following DBS in the immediate postoperative period, with limited data on long-term trajectories¹¹. Although short-term readmission rates were slightly higher in the DBS group, the difference reversed over longer follow-up, with progressively fewer readmissions observed at 3 and 5 years. This pattern may reflect a stabilization of disease course and reduced complication-driven hospitalizations once postoperative recovery is complete. In this context, outcomes such as institutionalization, mortality, and total healthcare expenditures may represent more stable indicators of long-term impact than readmission.

Despite these limitations, the outcomes appeared broadly consistent across hospital types, and post-discharge readmissions as well as total costs did not materially differ between teaching and non-teaching institutions. This uniformity argues against the notion that favorable trajectories are confined to highly specialized centers and may suggest the external generalizability of DBS^30,31.

In the full database analysis, DBS utilization clustered among younger, male patients treated in urban, high-volume teaching hospitals. Race, ethnicity, insurance and region also had an impact on access. These patterns are supportive of prior syntheses showing systematically lower DBS uptake among women and Black or Hispanic patients, even after adjustment for clinical factors and across Medicare, as well as other U.S. cohorts^30,31. Such disparities are unlikely to be explained solely by differences in disease severity; rather, they likely reflect cumulative factors encountered along the multi-step pathway to surgery, patient awareness and preferences, primary to specialty referral, multidisciplinary candidacy evaluation, coverage and out-of-pocket costs, and geographic reach of specialist services^32–35. Our finding that post-discharge outcomes were broadly similar between teaching and non-teaching hospitals strengthens external validity: it suggests that once appropriately selected and delivered, DBS performance is reproducible across institutional types, and that upstream access barriers, not technical performance, are the dominant drivers of inequity. This interpretation aligns with reports that short-term surgical outcomes do not differ substantially across urban versus non-urban settings and with contemporary device/programming advances that may impact inter-facility variability^20,36. The persistent concentration of procedures in academic centers, therefore, may reflect structural constraints in referral networks, reimbursement design, and regional workforce distribution rather than an intrinsic need for highly specialized environments³¹. Several practical strategies may help to reduce inequities in DBS access. Early and standardized screening for surgical candidacy, streamlined referral pathways, and broader use of telemedicine can help minimize geographic barriers^37–39. Simplifying prior authorization and establishing transparent, equitable reimbursement systems would further improve accessibility^40,41. Consistent patient education and shared decision-making across demographic groups, along with education of first-point-of-contact providers such as general neurologists and primary care physicians, and the strategic distribution of movement disorder specialists beyond major academic centers, will also be essential^42–44. Advances in adaptive and autonomous DBS technologies may eventually lessen dependence on specialized follow-up and may further reinforce more uniform care delivery^45–48. Ultimately, our findings suggest that improving the pathway to DBS, rather than focusing solely on the place of DBS delivery, may be important for promoting more equitable access.

This study has several limitations related to its retrospective observational design. Despite rigorous adjustment, residual confounding, code-based misclassification, and the absence of PD-specific clinical measures, such as motor severity and symptom burden, may have influenced the results. Although overlap weighting was applied to balance measured covariates, comparability between the DBS and non-DBS groups may remain imperfect. The non-DBS cohort included a substantial proportion of patients hospitalized for acute medical conditions, whereas patients selected for DBS are typically healthier and have fewer contraindications to surgery. Examination of principal diagnoses confirmed that many index hospitalizations in the non-DBS group were related to acute conditions rather than PD itself. To address this imbalance, we conducted sensitivity analyses excluding patients with dementia and restricting the non-DBS cohort to non-emergency admissions, which yielded largely consistent results, particularly for institutionalization and healthcare costs. However, unmeasured factors such as mild cognitive impairment, caregiver support, access to specialized care, and outpatient follow-up frequency may still differ between groups. Therefore, this comparison does not emulate randomized treatment assignment but instead reflects differences between surgically selected patients and a heterogeneous hospitalized PD population. The observed associations should not be interpreted as causal effects of DBS but rather as reflecting patient selection, baseline health status, and differences in access to care. Cost analyses were restricted to patients with nonzero observed costs, which may introduce selection bias. In addition, administrative claims data have inherent limitations in outcome definitions: readmissions cannot be reliably distinguished as planned versus unplanned, and mortality may be under-ascertained if deaths occur outside captured encounters. To mitigate confounding, overlap weighting was applied using propensity scores incorporating all available covariates, and Firth penalized logistic regression was used for sparse outcomes. Gamma regression with a log link was applied to account for the right-skewed distribution of healthcare expenditures. Future research should aim to establish prospective registries that integrate clinical severity indices, device characteristics, and precise surgical timelines for reoperations and battery replacements to enable more refined causal inference. Updated cost-effectiveness frameworks that incorporate device longevity, reprogramming requirements, and replacement intervals, along with systematic evaluation of equity across demographic, regional, and insurance groups, will also be necessary.

This nationwide real-world study suggests that DBS treatment is associated with more favorable long-term outcomes in survival, independence, and healthcare utilization. Given the observational nature of this study, causal inferences cannot be made. Continued efforts to improve analytic precision and ensure equitable access will be important to better understand and optimize the role of DBS in routine clinical care.

Methods

Study design and data source

This retrospective cohort study utilized the PINC-AI™ Healthcare Database (Premier, Inc., Charlotte, NC), a large U.S. administrative claims database encompassing inpatient, outpatient, and facility encounters across geographically and demographically diverse hospitals^49–51. The database includes comprehensive clinical and billing information from acute care hospitals nationwide and captures approximately 30% of all U.S. hospitalizations annually, including patients covered by Medicare, Medicaid, and private insurance.

Study population

We identified adults aged ≥50 years with a confirmed diagnosis of PD (ICD-10-CM code G20) who were hospitalized between January 1 and December 31, 2018. Patients who underwent initial DBS implantation during the index hospitalization in 2018 were classified as the DBS group. For patients in the non-DBS group, the index date was defined as the discharge date of a PD–related hospitalization in 2018. Patients with any record of DBS implantation prior to or after 2018 were excluded from the analysis to ensure a clear comparison between newly implanted DBS patients and those without any history of DBS. The date of DBS hospitalization was defined as the index date and assigned to non-DBS comparators. Patients were followed for up to 5 years after discharge, with outcomes evaluated at 1, 3, and 5-year intervals. Patients were excluded if the index hospitalization exceeded 60 days, duplicate records were identified, or key demographic or clinical variables (sex, admission source, or discharge disposition) were missing. After applying these criteria, 482 patients who underwent DBS and 79,363 non-DBS group patients were included in the analytic cohort (Supplementary Fig. 1).

Exposure and outcomes

The exposure of interest was DBS implantation during the index hospitalization. Outcomes were evaluated at 1, 3, and 5 years following discharge. Primary outcomes included (1) hospital readmission, (2) institutionalization, (3) all-cause mortality, and (4) total medical costs. Readmission was defined as any inpatient encounter within the specified follow-up period. Institutionalization was defined as admission to a skilled nursing facility, intermediate care facility, or long-term care institution. Mortality was identified through death records from subsequent inpatient, outpatient, or facility claims. Total medical costs were defined as cumulative direct healthcare expenditures incurred during the follow-up period after discharge, including inpatient, outpatient, procedural, and long-term care services. Cost data were derived from hospital-reported cost accounting systems and included both fixed and variable components. Patients with zero observed costs during each follow-up period were excluded, as these likely reflect absence of recorded encounters or incomplete cost capture rather than true absence of healthcare utilization. An alternative analytic strategy would be the use of a two-part model to account for zero-inflated cost distributions; however, given the nature of zero-cost observations in this dataset, we restricted the analysis to patients with observed costs. Costs associated with the index hospitalization, including DBS implantation, were not included.

Covariates

Covariates included demographic, clinical, and hospital characteristics with potential to confound the association between DBS and outcomes: age, sex, race, marital status, dementia, Charlson Comorbidity Index (CCI), residence type (urban/rural), hospital teaching status, bed size, region, admission source, discharge disposition, and insurance type.

Statistical analysis

All analyses were performed using R (version 4.4). Propensity scores representing the probability of undergoing DBS were estimated using multivariable logistic regression including all covariates listed above. To adjust for baseline differences and emphasize the region of common support, overlap weighting based on propensity scores was applied. Each DBS patient was weighted by 1 − PS and each non-DBS patient by PS, where PS denotes the individual propensity score. Covariate balance was assessed using standardized mean differences (SMD < 0.1 indicating adequate balance) and confirmed visually. After weighting, the effective sample size, reflecting the precision of weighted estimates, was 474 for the DBS group and 5321 for the non-DBS group.

Weighted logistic regression models were used to estimate the odds of hospital readmission, institutionalization, and mortality at 1, 3, and 5 years after discharge. Firth penalized logistic regression was applied for outcomes with sparse events to reduce small-sample bias. For medical costs, weighted generalized linear models with a log link and gamma distribution were used to account for skewed cost data.

To further address potential differences in baseline clinical characteristics and reasons for hospitalization between groups, we conducted additional sensitivity analyses. First, we examined the principal diagnoses associated with index hospitalizations in the non-DBS cohort using unweighted data to characterize the underlying reasons for admission. Non-emergency admissions were defined as hospitalizations not coded as emergency based on the admission type variable. Second, we performed sensitivity analyses excluding patients with dementia. Third, we restricted the non-DBS cohort to non-emergency admissions to reduce the influence of acute illness severity. All sensitivity analyses were conducted using the same propensity score weighting framework and modeling approaches as in the primary analyses.

As a supplementary analysis, we compared outcomes according to whether DBS implantation performed at teaching or non-teaching hospitals. Post-discharge hospital readmission and total medical costs at 1, 3, and 5 years were evaluated. Readmission was analyzed using logistic regression models, and costs were examined using generalized linear models with a log link and gamma distribution. Institutionalization and mortality were excluded from this sub-analysis because of the limited number of events.

To identify factors associated with DBS utilization, we performed a multivariable logistic regression analysis including all DBS recipients captured in the PINC-AI™ Healthcare Database. The dependent variable was DBS implantation, and independent variables included age, sex, race, Charlson Comorbidity Index, insurance type, hospital teaching status, urban/rural location, and U.S. Census region. Adjusted odds ratios (aORs) and 95% confidence intervals (CIs) were calculated. Continuous variables were summarized as medians with interquartile ranges, and categorical variables as frequencies and percentages. All tests were two-sided, with p < 0.05 considered statistically significant.

Ethical considerations and reporting

This study used de-identified administrative claims data from the PINC-AI™ Healthcare Database and was classified as non-human subjects research; therefore, institutional review board approval and informed consent were waived in accordance with applicable regulations governing research using anonymized datasets. The study was conducted in accordance with the Declaration of Helsinki and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) and the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) guidelines.

Author contributions

H.K. conceived and designed the study, organized and executed the project, performed the statistical analyses, and drafted the manuscript. R.M.B., T.R.M., and M.R. contributed to manuscript review and critical revision. A.B. and A.S. contributed to project execution and manuscript review. M.S.O. contributed to study conception and manuscript revision. C.W.H. conceived and organized the study, supervised the project, contributed to statistical analysis and interpretation, and critically revised the manuscript. All authors reviewed and approved the final manuscript.

Data availability

The data that support the findings of this study are available from the PINC-AI™ Healthcare Database (Premier, Inc.,Charlotte, NC), a proprietary administrative claims dataset. Restrictions apply to the availability of these data, which were used under license for the current study and are not publicly available. Access to the data may be obtained directly from Premier, Inc., subject to a data use agreement.

Code availability

Analytic code and non-sensitive variable definitions are available from the authors upon reasonable request.

Competing interests

H.K. received grant support from the Japanese Society of Neurology and the Melvin Yahr Clinical Research Award from the IAPRD/Parkinson’s Foundation, unrelated to the present study. R.M.B.: None. T.R.M.: None. M.R.: None. A.B.: Employment (Parkinson’s Foundation). A.S.: Employment (Parkinson’s Foundation). M.S.O.: Okun serves as Medical Advisor to the Parkinson’s Foundation and has received research grants from the NIH, Parkinson’s Foundation, the Michael J. Fox Foundation, the Parkinson Alliance, the Smallwood Foundation, the Tourette Association of America, and the UF Foundation. Dr. Okun’s research is supported by NIH awards R01NS131342, R01NR014852, R01NS096008, UH3NS119844, and U01NS119562; he is a multi-PI of the NIH R25NS108939 training grant. Okun has received royalties for publications with Hachette Book Group, Demos, Manson, Amazon, Smashwords, Books4Patients, Perseus, Robert Rose, Oxford, and Cambridge (movement disorders books). He is an Associate Editor for NEJM Journal Watch Neurology and JAMA Neurology. Okun has participated in CME and educational activities (past 12–24 months) on movement disorders sponsored by WebMD/Medscape, RMEI Medical Education, the American Academy of Neurology, the International Parkinson and Movement Disorder Society, Mediflix, and Vanderbilt University. The institution, and not Okun, receives grants from industry. He has participated as site PI and/or co-I in several NIH, foundation, and industry-sponsored trials without receiving honoraria. Research projects at the University of Florida receive device and drug donations. C.W.H.: C.W.H. has served as an Independent Safety Advisor for Saluda Medical and has received research funding from the Parkinson’s Foundation and honoraria for speaking engagements, grant reviews, and scientific reviews from The Parkinson’s Foundation, World Parkinson Congress, and UpToDate. C.W.H.: CWH has served as an Independent Safety Advisor for Saluda Medical and has received research funding from the Parkinson’s Foundation and honoraria for speaking engagements, grant reviews, and scientific reviews from The Parkinson’s Foundation, World Parkinson Congress, and UpToDate.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41531-026-01445-1.