Sequential pathway analysis of sex differences in deep brain stimulation for Parkinson’s disease

Stage 1: Referral

Patients with idiopathic, levodopa-responsive PD with optimized medication are typically referred for DBS consultation when they develop one or more of the following: troublesome motor fluctuations, levodopa-induced dyskinesias, or medication-refractory tremors.

Stage 2: Multidisciplinary evaluation

A multidisciplinary assessment is conducted, including a neuropsychological evaluation and a review of brain imaging to confirm suitability for electrode placement. Physical, occupational, and speech evaluations assess the patient’s mobility, balance, communication, airway protection (swallowing), and function. Their insights help predict the likelihood of patient recovery and benefit from DBS. A social worker assesses the patient’s support system, access to care, and capacity to attend the multiple postoperative appointments necessary to optimize stimulator settings and identify potential obstacles that could affect long-term outcomes.

Stage 3: Recommendation

A patient is ultimately approved/recommended for DBS for PD after 1) evaluation confirms they meet clinical and functional criteria, 2) they agree to cooperate during the procedure and follow-up, and 3) have a reliable support system. ¹⁴ Institutions performing DBS must have experienced neurosurgeons, appropriate imaging and surgical equipment, and a team trained in movement disorders.

Stage 4: Receipt

Patients who ultimately receive DBS are those with idiopathic PD who experience significant motor complications—such as dyskinesias, tremor, or motor fluctuations—that are inadequately controlled by medication. A robust and sustained levodopa response is the most consistent predictor of favorable DBS outcomes, and many centers require a formal levodopa challenge test to confirm this responsiveness. ¹⁵ Ideal candidates are typically younger than 70, healthy, have minimal cognitive impairment, and present stable psychiatric health.

This process involves multiple levels of influence by both individuals with PD and their providers that potentially contribute to observed sex differences in DBS surgery. To explore this multi-stage process, data from a large single-center cohort of patients traversing a standardized, multidisciplinary DBS pathway at a multidisciplinary, neurological care institute between 2012 and 2024 were analyzed to estimate sex differences in the conditional likelihood of progression to each stage—referral, evaluation, recommendation, and receipt—adjusting for clinical and demographic heterogeneity.

Data

This study utilized information collected in the INFORM (INstitute FOR Neurological diseases) research database, a longitudinal, clinical research database housing data on nearly 20,000 patients—maintained by the Norman Fixel Institute for Neurological Diseases (Fixel Institute) at the University of Florida (UF) Health in Gainesville, Florida. The Fixel Institute—a recognized Parkinson’s Foundation Center of Excellence—is an interdisciplinary hub offering diagnosis, advanced treatment, education, and research, providing neurology, neurosurgery, neuropsychology, psychiatry, rehabilitative therapies, genetics counseling, nutrition, and support services.

The INFORM database includes over 7,500 patients diagnosed with idiopathic PD, 2,000 with dystonia, and more with other movement disorders. Patients who attend at least one medical appointment may join the database, thereby allowing their information to be used for research purposes. The database captures standardized clinical assessments encompassing motor function, mood, cognition, medication history, and neurosurgical interventions, including DBS, referral status, pre-surgical evaluations, approval decisions, and surgical outcomes, as well as device programming and follow-up assessments.

Sample

Inclusion criteria: The study included unique adult patients with PD who were seen at the Fixel Institute between February 2012 and February 2024 and enrolled in the INFORM research database. Eligible participants had documented clinical information relevant to progression through the DBS pathway, including referral, multidisciplinary evaluation, recommendation, approval, and, when applicable, receipt of DBS. The analytic sample was limited to patients with idiopathic PD and available demographic and clinical data for stage-specific analyses. Exclusion criteria: Patients were excluded if they were not unique individuals, were seen outside the study period, did not have PD, or lacked sufficient clinical information to determine progression through the DBS pathway. Because each stage-specific model used complete-case analysis, individuals with missing data on variables required for a given stage were excluded. Patients with non-idiopathic PD or other movement disorders were not included in the analytic cohort.

These criterion results in 4,310 unique patients. Patient information included age, sex, race, insurance, marital status, comorbidities, and PD duration, type, and medication. Clinical PD treatment information included whether the patient was referred for multi-disciplinary evaluation, counseled for DBS surgery, received multidisciplinary evaluation, recommended for DBS, approved for DBS, or received DBS. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Sample size justification

Because this was a retrospective study using an existing clinical database, no formal a priori sample size calculation was performed. Instead, the analytic sample included all eligible unique patients with PD seen at the Fixel Institute between February 2012 and February 2024 who met study inclusion criteria. This yielded a total sample of 4,308 individuals. To ensure stable estimation in the stage-specific logistic regression models, the number of predictors included at each stage was limited according to an events-per-parameter (EPV) rule targeting approximately 10 events per parameter, an approach commonly used to support reliable inference in logistic regression with relatively rare outcomes. Given 186 referral events in Stage 1 and progressively smaller risk sets in subsequent stages, this approach provided the basis for determining the feasible complexity of the models and justified the use of the full available cohort.

Characteristics

Characteristics were chosen a priori from available factors: age, marital status, idiopathic PD, insurance status, sex, problematic tremors, levodopa use, comorbidity burden, and race/ethnicity. To improve numerical stability, continuous predictors were standardized (z-scored) within stage to reduce collinearity and scaling issues^16,17 and categorical predictors with fewer than five observations were collapsed. Only type 1 diabetes mellitus (DMT1), type 2 diabetes mellitus (DMT2), hypertension, stroke, cardiovascular disease (CVD), cancer (non-skin), and respiratory disease comorbidities were available. Therefore, the comorbidity burden represented the sum of the number of these diagnoses reported by the patient. Patients without any of these diagnoses were assigned a value of 0. Each stage used complete-case analysis for the variables included in that stage’s model.

DBS pathways

Progression along the sequential DBS pathways was represented as binary stages: (S1) referral for DBS; (S2) receipt of a multidisciplinary evaluation among those referred; (S3) recommendation for DBS among those evaluated; and (S4) receipt of DBS among those recommended. Since the probability of advancing to each stage was conditional on success at the previous stage, each stage was modeled using stage-specific logistic regression. Since stages S2 through S4 were imbalanced (very few non-advancers), the predictors per stage were limited using an events-per-parameter (EPV) rule (targeting EPV=10)—a standard shown in prior work to yield stable estimates. ¹⁸

Analytic framework and statistical analysis

The pathway was represented as four sequential stage-specific logistic regressions—(S1): referral for DBS, (S2): receipt of a multidisciplinary evaluation among those referred, (S3): recommendation for DBS among those evaluated, and (S4): receipt of DBS among those approved. To address low cell counts and differential success rates between men and women, a penalized likelihood (Firth) logistic regression was used as the default estimator, with predefined fallbacks described below. For each stage s∈{1,…,4}, Y_s denotes the binary outcome for the risk set eligible for stage s. The model included sex, all main effects for the characteristics (X), and sex-by-predictor interactions (see Equation (1)).

l o g i t { P f ( Y s = 1 ) } = σ s + β s ( s e x ) I ( F e m a l e ) + ∑ j γ s j X j + ∑ j δ s j [ X j · I ( F e m a l e ) ]

(1)

Since full interaction blocks can be non-estimable when events are rare, a two-step procedure was used. In Stage 1, each predictor X_i, was checked to see whether its effect differed between men and women, using a method that handles rare events (Firth regression). A simple, or ‘Base’, model (Equation (2a)) was fit that included sex and one factor at a time. Next, a more detailed ‘Interaction’ (Equation (2b)) model was tested, which allowed the factor’s effect to differ by sex.

Stage 1: One-at-a-time interaction scan

B a s e : l o g i t { Pr ( Y s = 1 ) } = σ + β ( s e x ) I ( F e m a l e ) + γ j X j

(2a)

I n t e r a c t i o n : l o g i t { Pr ( Y s = 1 ) } = σ + β ( s e x ) I ( F e m a l e ) + γ j X j + δ j [ X j · I ( F e m a l e ) ]

(2b)

After both the Stage 1 base and interaction models converged via Firth regression, a penalized likelihood ratio test (PLRT) was performed to evaluate whether the extra interaction terms improved model fit. If the one-at-a-time interaction was non-estimable by Firth, a mean bias-reduced logistic regression was conducted, and a Wald block test was performed on the vector of interaction coefficients. Predictors with estimable interactions using either method were ‘accepted.’

Stage 2: Reduced form full interaction model

Using the full set of ‘accepted’ predictors from Stage 1, Stage 2 estimated a reduced-form interaction model that included sex, main effects for the candidate predictors, and ‘accepted’ sex-by-predictor interactions of the form in Equation (3).

l o g i t { P f ( Y s = 1 ) } = σ + β ( s e x ) I ( F e m a l e ) + ∑ j γ j X j + ∑ j ϵ J O K δ j [ X j · I ( F e m a l e ) ]

(3)

If the penalized fit failed, mean-bias-reduced estimation was used. Sex-specific odds ratios (ORs) and associated 95% confidence intervals (CIs) were reported. Odds ratios (ORs) for continuous variables are reported per 1 SD increase. ¹⁹ For predictors that did not interact with sex in the reduced model, the coefficients applied equally to both sexes. For predictors with sex interactions, let δ j ^ be the coefficient for the interaction with predictor X _j and Female. Since male is the reference category, the male OR is exp ( γ j ^ ) and the female OR is exp ( γ j ^ + δ j ^ ) .

Because several stage-specific risk sets were small and imbalanced, safeguards were implemented to obtain valid estimates despite sparse cells. First, Firth’s penalized likelihood logistic regression was used as the default estimator. Firth’s correction reduces small-sample bias and is robust when standard maximum likelihood. ²⁰ To improve stability, the number of iterations was increased to 1,000, and the maximum step size was decreased to 0.2. If Firth’s regression failed to converge, the model was respecified as a mean-bias-reduced logistic regression to obtain consistent estimates. This fallback was automatic, ensuring that differences in estimates reflected the fitting method rather than covariate adjustments. Penalized estimation and bias-reduced fallback, combined with principled level consolidation, within-stage predictor standardization, and disclosure of risk-set sizes. All analyses were conducted in R ²¹ using these packages: logistf ²² for Firth penalized logistic regression, brglm2 ²³ for mean bias-reduced logistic regression, and tidyverse ²⁴ for data consolidation and output.

Robustness and sensitivity analyses

Three additional analyses were performed to evaluate the robustness of the results. First, sex-stratified Firth logistic models were fit to ensure that removing non-estimable interactions did not mask clinically relevant differences. These models provided sex-specific ORs that did not rely on any explicit interaction terms and corroborated the direction and approximate magnitude of the effects observed in the pooled reduced-interaction analysis. Second, the estimation was repeated using mean-bias-reduced logistic regression as the primary estimation method to validate the approach. Finally, the EPV cap was lowered to five to test the sensitivity to this limitation.

This study has been approved by the University of Florida Institutional Review Board (#IRB202300118).

Sample characteristics

Of the 4,308 individuals (Table 1) in the analytic cohort (S1), two-thirds were male (2,866; 66.5%), and one-third were female (1,442; 33.5%). The higher proportion of males in the cohort (66.5%) is consistent with the known higher prevalence of PD among men and reflects expected epidemiologic patterns of the underlying clinical population rather than sampling bias. Only 186 patients (≈4.3% of the cohort) were referred for DBS (S2), and the sex distribution at referral matched that of the cohort: 125 males (67.2%) and 61 females (32.8%). Among those referred, 180 (97%) completed a multidisciplinary evaluation (S3), again with 121 males (67.2%) and 59 females (32.8%). Among the patients evaluated, 159 (88%) underwent DBS surgery (S4): 105 males (66.0%) and 54 females (34.0%). Overall, most attrition occurred before referral, and the male–female composition remained largely stable across stages, with a slight uptick in the female share at surgery (32.8% at S2/S3 to 34.0% at S4).

OPEN IN VIEWER

Table 1.

Sample size and characteristics by sex, stage.

​	Male
	N (%)	Age		Comorbidities		Tremors		Idiopathic		Married		No insurance		Levodopa		Non-white
		Mean	SD	Mean	SD	N	PCT	N	PCT	N	PCT	N	PCT	N	PCT	N	PCT
S1	2866 (66.50%)	74.4	10.21	0.26	0.79	6	0.20%	2862	99.90%	295	68.90%	17	0.60%	371	12.90%	2559	89.30%
S2	125 (67.20%)	68.83	9.51	1.86	1.19	1	0.80%	125	100.00%	88	71.00%	4	3.20%	114	91.20%	33	26.40%
S3	121 (67.20%)	68.57	9.48	1.82	1.14	1	0.80%	121	100.00%	86	71.70%	4	3.30%	110	90.90%	33	27.30%
S4	105 (66.00%)	67.72	9.24	1.89	1.17	1	1.00%	105	100.00%	76	73.10%	3	2.90%	95	90.50%	25	23.80%

​	Female
	N	Age		Comorbidities		Tremors		Idiopathic		Married		No insurance		Levodopa		Non-white
		Mean	SD	Mean	SD	N	PCT	N	PCT	N	PCT	N	PCT	N	PCT	N	PCT
S1	1442 (33.50%)	74.52	10.15	0.27	0.79	4	0.30%	1434	99.40%	144	61.30%	8	0.60%	201	13.90%	1287	89.30%
S2	61 (32.80%)	70.44	8.54	1.51	0.96	1	1.60%	60	98.40%	42	68.90%	2	3.30%	52	85.20%	17	27.90%
S3	59 (32.80%)	70.05	8.39	1.54	0.95	1	1.70%	59	100.00%	40	67.80%	2	3.40%	50	84.70%	17	28.80%
S4	54 (34.00%)	69.76	8.39	1.59	0.96	1	1.90%	54	100.00%	38	70.40%	2	3.70%	45	83.30%	12	22.20%

As seen in Table 1, among males, the mean age decreased from 74.4 (SD=10.21) years in the full cohort (S1, N=2,866) to 68.83 (SD=9.51) at referral (S2, N=125), 68.57 (SD=9.48) at evaluation (S3, N=121), and 67.72 (SD=9.24) at surgery/approval (S4, N=105). Average comorbidity burden rose from 0.26 (SD=0.79) at S1 to 1.86 (SD=1.19) at S2, 1.82 (SD=1.14) at S3, and 1.89 (SD=1.17) at S4. Tremor documentation was rare at baseline (0.2%) and remained uncommon (≈1%) in later stages. Nearly all men had idiopathic PD throughout (≥99.9%). The proportion married was consistently high, increasing slightly from 68.9% (S1) to 73.1% (S4). Lack of insurance was uncommon overall (0.6% at S1) but somewhat higher among those referred/evaluated (∼3% at S2–S4). Levodopa use increased dramatically from 12.9% at baseline to about 91% among referred/evaluated groups, remaining stable thereafter. The share classified as non-White fell sharply from 89.3% at S1 to 26–27% at S2–S3 and 23.8% at S4.

Among females, mean age decreased from 74.52 (SD=10.15) years at S1 (N=1,442) to 70.44 (SD=8.54) at S2 (N=61), 70.05 (SD=8.39) at S3 (N=59), and 69.76 (SD=8.39) at S4 (N=54). Comorbidities rose from 0.27 (SD=0.79) at S1 to 1.51 (SD=0.96) at S2, 1.54 (SD=0.95) at S3, and 1.59 (SD=0.96) at S4. Tremor documentation was rare but slightly more frequent than in males at later stages (≈1.6–1.9%). Idiopathic PD remained nearly universal (≥98–100%). Marriage rates were lower than in males at baseline (61.3%) but converged across stages (70.4% at S4). Uninsured status was uncommon (0.6% at S1) and modestly higher among referred/evaluated (∼3–4%). Levodopa use showed the same step-up pattern (from 13.9% at S1 to ∼85% at S2–S4). As with males, the proportion of non-White decreased with progression (from 89.3% at S1 to ∼28% at S2–S3 and 22.2% at S4).

Sex-only adjusted model

Adjusting only for sex, the Firth estimation (Table 2) showed no statistically significant sex differences at any stage of the DBS pathway. At referral (S1), females had odds like males (OR = 0.97, CI=0.71, 1.32; p = 0.86; 186 events among 4,308). Among those referred, the odds of completing a multidisciplinary evaluation (S2) were also comparable (OR = 0.88, CI=0.19, 5.17; p = 0.88; 180 events among 186), though the confidence interval was wide. Conditional on evaluation, females showed higher—but imprecisely estimated—odds of being recommended for DBS (S3) (OR = 1.55, CI=0.59, 4.70; p = 0.39; 159 events among 180). Finally, among those recommended, females had lower—again imprecise—odds of receiving surgery (S4) (OR = 0.61, CI=0.17, 2.38; p = 0.46; 150 events among 159). The progressively smaller risk sets at later stages produce wide confidence intervals, so these stage-specific trends should be interpreted cautiously.

OPEN IN VIEWER

Table 2.

Sex-adjusted penalized likelihood model estimation.

Stage	N	Events	Odds ratio	95% confidence interval		p-value
S1: Referral	4308	186	0.972	0.708	1.32	0.86
S2: Multidisciplinary Evaluation | S1	186	180	0.881	0.19	5.17	0.876
S3: Recommended | S2	180	159	1.55	0.59	4.7	0.387
S4: DBS Surgery | S3	159	150	0.614	0.167	2.38	0.464

Fully adjusted model

(S1) As seen in Table 3, among men, older age was associated with lower odds of referral (OR = 0.95, CI=0.93, 0.97, p<0.001), and greater comorbidity burden with higher odds (OR = 1.19, CI=1.00, 1.41, p=0.05). No other male predictors were significant. Among women, being unmarried was linked to lower odds of referral (OR = 0.71, CI=0.38, 0.92, p=0.03); other factors were not significant, though age trended inversely (OR = 0.98, CI=0.95, 1.00, p=0.07).

OPEN IN VIEWER

Table 3.

Adjusted penalized likelihood model estimation.

Stage	Predictor	Male				Female
		Odds ratio	95% CI		p-value	Odds ratio	95% CI		p-value
S1: Referral	Age	0.95	0.93	0.97	0.00	0.98	0.95	1.00	0.07
S1: Referral	Tremors	0.82	0.08	4.29	0.83	1.03	0.15	7.25	0.98
S1: Referral	Idiopathic	1.48	0.07	225.00	0.81	0.83	0.10	6.85	0.86
S1: Referral	Unmarried	0.95	0.58	1.52	0.82	0.71	0.38	0.92	0.03
S1: Referral	No Insurance	0.65	0.18	1.93	0.45	0.81	0.17	3.87	0.79
S1: Referral	Levodopa	1.49	0.76	3.13	0.26	0.96	0.42	2.20	0.92
S1: Referral	Non-White	0.84	0.51	1.36	0.48	0.76	0.40	1.45	0.40
S1: Referral	Comorbidities	1.19	1.00	1.41	0.05	0.92	0.72	1.18	0.52
S2: Multidisciplinary Evaluation | S1	Age	0.93	0.79	1.03	0.19	0.89	0.77	1.03	0.11
S2: Multidisciplinary Evaluation | S1	Tremors	0.03	0.00	7.79	0.17	0.08	0.00	9.11	0.29
S2: Multidisciplinary Evaluation | S1	Idiopathic	9.59	0.07	298.00	0.32	9.59	0.07	298.00	0.32
S2: Multidisciplinary Evaluation | S1	Unmarried	0.23	0.02	1.78	0.15	1.57	0.09	26.60	0.76
S2: Multidisciplinary Evaluation | S1	No Insurance	0.12	0.00	27.80	0.36	0.19	0.00	12.90	0.44
S2: Multidisciplinary Evaluation | S1	Levodopa	1.22	0.01	22.30	0.91	1.24	1.06	1.90	0.05
S2: Multidisciplinary Evaluation | S1	Non-White	5.71	0.38	248.00	0.27	1.00	0.05	20.90	1.00
S2: Multidisciplinary Evaluation | S1	Comorbidities	0.51	0.20	0.87	0.05	1.88	0.54	6.61	0.32
S3: Recommended | S2	Age	0.91	0.84	0.97	0.00	0.91	0.79	1.04	0.17
S3: Recommended | S2	Tremors	0.19	0.00	36.70	0.43	2.64	0.07	102.00	0.60
S3: Recommended | S2	Unmarried	0.70	0.21	2.54	0.57	1.95	0.21	18.40	0.56
S3: Recommended | S2	No Insurance	0.61	0.05	16.00	0.75	5.62	0.18	172.00	0.32
S3: Recommended | S2	Levodopa	0.89	0.07	5.54	0.91	2.42	0.07	88.60	0.63
S3: Recommended | S2	Non-White	0.29	0.09	0.88	0.03	0.01	0.00	0.31	0.01
S3: Recommended | S2	Comorbidities	1.90	1.09	3.73	0.02	4.20	0.89	19.90	0.07
S4: DBS Surgery | S3	Age	0.94	0.81	1.04	0.25	1.02	0.90	1.14	0.80
S4: DBS Surgery | S3	Tremors	0.06	0.00	18.50	0.28	1.55	1.04	1.87	0.05
S4: DBS Surgery | S3	Unmarried	0.95	0.16	9.04	0.96	0.36	0.05	2.43	0.29
S4: DBS Surgery | S3	No Insurance	0.32	0.02	49.70	0.54	0.49	0.02	13.30	0.67
S4: DBS Surgery | S3	Levodopa	0.48	0.00	6.46	0.65	0.82	0.05	14.60	0.89
S4: DBS Surgery | S3	Non-White	0.15	0.02	0.91	0.04	0.12	0.01	1.18	0.07
S4: DBS Surgery | S3	Comorbidities	1.87	0.79	5.63	0.16	3.13	0.86	11.50	0.08

(S2) In men, more comorbidities predicted lower odds of completing evaluation (OR = 0.51, CI=0.20, 0.87, p=0.05). In women, levodopa therapy was associated with higher odds of evaluation (OR = 1.24, CI=1.06, 1.90, p=0.05). All other effects were imprecise with wide CIs at this stage.

(S3) For men, older age again predicted lower odds of recommendation (OR = 0.91, CI=0.84, 0.97, p<0.01), non-White race predicted lower odds (OR = 0.29, CI=0.09, 0.88, p=0.03), and comorbidities predicted higher odds (OR = 1.90, CI=1.09, 3.73, p=0.02). For women, non-White race was strongly associated with lower odds of recommendation (OR = 0.01, CI=0.01, 0.31, p=0.01); other predictors were not significant.

(S4) Among men, non-White race was associated with lower odds of surgery (OR = 0.15, CI=0.02, 0.91, p=0.04). Among women, tremor predicted higher odds of proceeding to surgery (OR = 1.55, CI=1.04, 1.87, p=0.05). Other effects at S4 were not statistically significant.

Robustness and sensitivity analyses

Appendix Table 1 presents the results of the sex-stratified first logistic regression models. (S1) Among men, older age was linked to lower odds of referral (OR = 0.95, CI=0.93, 0.97, p<0.001); higher comorbidity burden showed a borderline increase in referral (OR = 1.19, CI=1.00, 1.41, p=0.05). Among women, age was also inversely associated with referral (OR = 0.97, CI=0.94, 0.99, p=0.02). In addition, being unmarried was associated with lower referral odds (OR = 0.27, 0.13–0.55, p<0.001), while levodopa therapy was strongly associated with higher referral odds (OR = 13.60, CI=5.33, 36.91, p<0.001). (S2) Among men, comorbidities predicted lower odds of completing evaluation (OR = 0.51, CI=0.20, 0.91, p=0.05). In women, levodopa was associated with higher odds of evaluation (OR = 2.43, CI=2.20, 2.52, p=0.05). (S3) For men, older age again reduced the odds of recommendation (OR = 0.91, CI=0.84, 0.97, p<0.01), non-White race was associated with lower odds (OR = 0.29, CI=0.09, 0.88, p=0.03), and comorbidities increased the odds (OR = 1.90, CI=1.09, 3.73, p=0.02). For women, non-White race showed a marked reduction in recommendation (OR = 0.02, CI=0.01, 0.25, p<0.001). (S4) Among men, non-White race remained associated with lower odds of surgery (OR = 0.15, CI=0.02, 0.91, p=0.04). Among women, tremor predicted higher odds of proceeding to surgery (OR = 1.11, CI=1.01, 1.21, p=0.05), and non-White race again indicated lower odds (OR = 0.09, CI=0.02, 0.13, p=0.05).

Appendix Table 2 presents the results of the sensitivity analysis, using mean-bias-reduced logistic regression as the primary estimation method. (S1) Among men, higher age was associated with lower odds of referral (OR = 0.95, CI=0.93, 0.97, p<0.001). Greater comorbidity burden showed a small, borderline increase in referral likelihood (OR = 1.19, CI=1.00, 1.41, p=0.05). (S2) In men, higher comorbidity burden was associated with lower odds of advancing to evaluation (OR = 0.51, CI=0.25, 1.00, p=0.05). In women, no predictors were significant. (S3) For men, older age again predicted lower odds of recommendation (OR = 0.91, CI=0.86, 0.97, p=0.01). Non-White men were less likely to be recommended (OR = 0.29, CI=0.10, 0.85, p=0.02), whereas greater comorbidity burden increased the odds (OR = 1.90, CI=1.07, 3.38, p=0.03). For women, race was strongly associated with recommendation; Non-White women had markedly lower odds (OR = 0.01, CI=0.00, 0.31, p=0.01). (S4) Among men, Non-White race remained associated with lower odds of receiving DBS (OR = 0.15, CI=0.03, 0.76, p=0.02). Among women, no predictors were statistically significant.

Appendix Table 3 presents the results of the final sensitivity analysis, in which the EPV was set to 5. These results differ little from those presented above.

Limitations

Seven methodological and conceptual limitations should merit consideration when interpreting these findings. First, although the INFORM database offers detailed, prospectively collected information on patients progressing through the DBS pathway, it represents a single, high-volume tertiary center in Florida; therefore, findings may not generalize to other institutions with different referral networks, demographic compositions, or multidisciplinary workflows. Second, the database does not capture several clinical indicators that may influence candidacy—such as disease severity, motor fluctuation profiles, neuropsychological test scores, symptom duration, or levodopa challenge responses—limiting the study’s ability to account for all relevant determinants of referral, evaluation, and surgical recommendation. Third, the operationalization of “race” as White versus non-White and reliance on a limited set of comorbidities restricts the conceptual richness of social and clinical heterogeneity that may shape DBS decision-making; similarly, unmeasured psychosocial factors (e.g., caregiving burden, treatment preferences, social support) that are known to differ by sex could not be evaluated. Fourth, this study was conducted at a single, high-volume tertiary care center with a standardized, multidisciplinary DBS pathway, which may limit the generalizability of findings to other clinical settings with differing referral patterns, patient populations, and care structures. Future multicenter studies applying similar stage-specific analytical approaches across diverse institutions are needed to validate these findings and enhance their external validity, potentially advancing understanding of disparities in DBS access. Fifth, although the sequential logistic modeling strategy allows for stage-specific estimation, the small risk sets at later stages yielded sparse data structures, necessitating the use of penalized likelihood (Firth) and mean-bias-reduced logistic regression. Sixth, while these approaches mitigate separation and small-sample bias, they result in wide confidence intervals, reduced statistical power, and limited precision for subgroup effects and interaction terms. Finally, the conditional modeling framework assumes measured and appropriately specified confounders at each stage; if unobserved factors influence both advancement through the pathway and sex differences in care, residual confounding may persist. Together, these limitations suggest caution when interpreting stage-specific estimates and highlight the need for multicenter, mixed-methods research that incorporates richer clinical, psychosocial, and structural determinants of DBS access.