Tinnitus constancy and self-reported severe headache or migraine symptoms in peri- and postmenopausal women: a statistical modeling and machine learning analysis Authors: Li Jiang, Ying Cui, Huimin Du Published: 2025 ABSTRACT AbstractBackgroundSevere headache or migraine symptoms are a highly prevalent and disabling neurological disorder with distinct clinical and mechanistic features in peri- and postmenopausal women. As research increasingly explores its complex sensory phenotype, auditory comorbidities such as tinnitus have drawn growing attention. Among them, the temporal constancy of tinnitus may be linked to severe headache or migraine symptom occurrence, yet remains understudied in this population.MethodsA total of 2,485 peri- and postmenopausal women from a nationally representative U.S. sample were included in this observational analysis. The target variable was severe headache or migraine symptoms, defined using standardized symptom and medication information. The main explanatory variable was tinnitus constancy, derived from standardized questionnaire items. Weighted multivariable logistic regression models were used to estimate the association between tinnitus constancy and severe headache or migraine symptoms, adjusting for demographic, socioeconomic, and clinical factors. Variable importance was further assessed using five complementary approaches: LASSO regression, logistic regression, random forest classification, the Boruta algorithm, and SHAP-based LightGBM modeling.ResultsA significant and graded association was observed between tinnitus constancy and severe headache or migraine symptoms. Compared to women without tinnitus, those with occasional (OR = 2.29, 95% CI 1.44–3.63), intermittent (OR = 2.62, 95% CI 1.57–4.35), and constant tinnitus (OR = 1.93, 95% CI 1.26–2.95) had higher odds of reporting severe headache or migraine symptoms. Intermittent tinnitus displayed the strongest association. Age was negatively associated with symptom reporting and consistently ranked as the most important predictor across all machine learning algorithms. SHAP dependence plots confirmed elevated symptom probability among younger individuals and those with more variable tinnitus patterns.ConclusionsTinnitus constancy, particularly in its intermittent form, is independently associated with higher odds of severe headache or migraine symptoms in peri- and postmenopausal women. Age was also a dominant protective factor. These findings highlight the importance of considering the temporal patterns of tinnitus symptoms and age-related vulnerability when evaluating severe headache or migraine symptoms in peri- and postmenopausal women.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40001-025-03712-y.Keywords: Severe headache or migraine symptoms, Perimenopausal women, Postmenopausal women, Tinnitus constancy, Machine learning Background Migraine is a highly prevalent neurological disorder [1, 2], affecting over one billion people and accounting for the second leading cause of years lived with disability worldwide[3]. Women experience disproportionately higher symptom burden[4, 5], partly due to fluctuations in sex hormones—especially estrogen—which influence sensory processing and pain susceptibility [6, 7]. These hormonal dynamics become particularly pronounced during the menopausal transition, during which some women report worsening symptoms, whereas others experience improvement once hormonal levels stabilize [8–12]. Peri- and postmenopausal women therefore represent a mechanistically distinct subgroup in whom sensory and neuroendocrine changes may alter headache-related symptom presentations [11, 13, 14]. As research continues to elucidate the multisystemic nature of migraine, increasing attention has turned toward sensory comorbidities that may modify migraine presentation or risk. These include photophobia, cutaneous allodynia, vestibular dysfunction, and auditory disturbances [15, 16]. Among the latter, tinnitus—defined as the perception of sound without an external source [17]—has emerged as a potential sensory correlate or modifier of migraine [18]. Shared neurophysiological mechanisms, such as thalamocortical dysrhythmia, central sensitization, and serotonergic dysfunction, have been proposed to underlie the co-occurrence of these conditions[19, 20]. Epidemiological studies suggest that individuals with migraine report tinnitus at higher rates than those without, and vice versa[21, 22]. Despite this observed comorbidity, most migraine research has focused on classical pain features and visual or auditory hypersensitivity, while auditory perception disturbances like tinnitus remain underexplored in headache medicine. A rarely examined dimension of tinnitus is its temporal pattern. Most epidemiologic studies classify tinnitus simply as present or absent[23], overlooking whether intermittent, occasional, or constant symptom courses may differentially relate to neurological outcomes. Conceptual models suggest that fluctuating tinnitus may provoke stronger limbic and attentional responses due to reduced habituation and heightened vigilance[24, 25], processes relevant to headache-related sensory dysfunction[26, 27]. In peri- and postmenopausal women, these dynamics may further interact with age- and hormone-related changes in pain modulation, mood regulation, and auditory sensitivity[28, 29]. However, no large-scale studies have evaluated whether tinnitus constancy is associated with self-reported severe headache or migraine symptoms, particularly in hormonally vulnerable midlife women. Accordingly, this study aimed to examine the association between tinnitus constancy and self-reported severe headache or migraine symptoms in a nationally representative sample of peri- and postmenopausal women. We further applied multivariable regression and machine learning-based feature-selection methods to identify key contributors to these symptoms, thereby clarifying the potential role of auditosry-symptom variability in headache-related presentations. Methods Study design and participants The National Health and Nutrition Examination Survey (NHANES) is a nationally representative surveillance program overseen by the U.S. National Center for Health Statistics (NCHS), aiming to assess health conditions within the civilian, non-institutionalized U.S. population. Data collection involves in-person household interviews combined with physical examinations, laboratory analyses, and nutritional assessments conducted in mobile examination centers. Approximately 5,000 individuals are selected each year through a complex, multistage probability sampling design to ensure representation across demographic and geographic subgroups. The survey protocol received ethical approval from the NCHS Institutional Review Board, and all participants completed written informed consent prior to participation [30]. Data were obtained from three consecutive cycles of the National Health and Nutrition Examination Survey (NHANES): 1999–2000, 2001–2002, and 2003–2004, selected based on the availability of detailed questionnaire items on severe headache or migraine symptoms during these periods. Because several headache-related questions were no longer collected in NHANES after 2004, these survey cycles are the most recent ones that retain the variables needed for classifying severe headache or migraine symptoms, consistent with previous NHANES-based analyses [31, 32]. The pooled dataset initially included 31,126 participants. To isolate a target population of peri- and postmenopausal women, male respondents (n = 15,184) and females who had not reached menopausal transition (n = 12,643) were excluded. Participants lacking data on severe headache/migraine symptom status (n = 1), tinnitus constancy (n = 5), or any key covariates (n = 808) were subsequently removed. The final analytical sample consisted of 2,485 women classified as peri- or postmenopausal. Identification of peri- and postmenopausal women Peri- and postmenopausal status was determined using the Reproductive Health Questionnaire data from NHANES. Female participants were initially asked whether they had experienced regular menstrual periods in the past 12 months, excluding bleeding caused by medical conditions or surgical procedures. Participants who reported an absence of regular menstruation were subsequently asked to indicate the reason for menstrual cessation. Participants who indicated “going-gone through menopause” as the reason for menstrual cessation were classified as peri- and postmenopausal and comprised the analytic sample for this study. Those who indicated current pregnancy, recent childbirth, breastfeeding, irregular periods, or medical conditions as reasons for amenorrhea were excluded from the analytical sample. Assessment of self-reported severe headache or migraine symptoms Self-reported severe headache or migraine symptoms were identified using two NHANES data components: the headache symptom questionnaire and prescription medication records. Symptom information originated from the Miscellaneous Pain questionnaire, in which respondents aged ≥ 20 years were asked whether they had experienced “severe headaches or migraines” during the preceding three months. Medication use was assessed through structured interviews documenting all prescription drugs taken within the previous 30 days. Because this item does not distinguish migraine from other primary or secondary headache disorders, it reflects a broad category of self-reported severe headache symptoms rather than a clinically validated migraine diagnosis. To classify medications, NHANES employs the Multum Lexicon Therapeutic Classification System, which organizes drugs into a multilevel hierarchy. Within this framework, “Antimigraine Agents” constitute the Level-3 therapeutic class (Class 193) under “Analgesics” (Level-2) and “Central Nervous System Agents” (Level-1). This class includes triptans, ergot derivatives, and other agents designated by Multum as indicated for migraine management. Identification of antimigraine medications followed this classification structure. Additional information regarding the Multum Lexicon database can be found at: https://www.cerner.com/solutions/drug-database. Participants were classified as having self-reported severe headache or migraine symptoms if they (1) answered “yes” to the questionnaire item on severe headaches or migraines, or (2) reported use of a medication belonging to the Multum “Antimigraine Agents” class. This definition captures symptomatic reports and treatment patterns but does not permit differentiation among headache subtypes or confirmation of a clinical migraine diagnosis. Accordingly, we interpret this variable as representing a self-reported severe headache/migraine symptom phenotype rather than definitive migraine. While this approach has been used in prior NHANES studies examining “severe headache or migraine” symptom patterns [31, 32], its application reflects the constraints of available survey items rather than a validated diagnostic standard. Assessment of tinnitus constancy Tinnitus experience was assessed using self-reported responses from the NHANES Audiometry Questionnaire. Participants aged 20 years or older were first asked whether they had experienced ringing, roaring, or buzzing in their ears within the past 12 months. Those who responded affirmatively were then asked to indicate the frequency of these symptoms. To capture the temporal pattern of symptom occurrence, tinnitus constancy—reflecting both the frequency and persistence of tinnitus—was categorized into three groups: constant tinnitus for those who reported symptoms occurring “almost always”; intermittent tinnitus for those indicating occurrence “at least once a day” or “at least once a week”; and occasional tinnitus for those reporting symptoms “at least once a month” or “less frequently than once a month”. This terminology has been used in prior tinnitus research to distinguish degrees of temporal persistence and to differentiate between constant and fluctuating tinnitus presentations [33]. Covariates Covariates were classified in accordance with the Social Determinants of Health (SDH) framework, which conceptualizes health outcomes as being shaped by both structural determinants and intermediary mechanisms [34]. This classification enabled a structured approach to model the multilevel influences contributing to severe headache or migraine symptoms among peri- and postmenopausal women. Structural determinants encompassed key demographic and socioeconomic characteristics. Age was treated as a continuous variable to account for its linear influence across the menopausal transition. Race and ethnicity were categorized into five groups: Mexican American, Non-Hispanic Black, Non-Hispanic White, Other Hispanic, and Other. Educational attainment was grouped as high school or below versus some college or higher, reflecting differential access to health-related knowledge and resources. Marital status was categorized as married or living with a partner, widowed/divorced/separated, and never married. To capture socioeconomic status, the poverty income ratio (PIR)—a continuous measure reflecting household income relative to the federal poverty level—was included. Intermediary determinants included lifestyle behaviors, biological indicators, and chronic disease status. Smoking and alcohol consumption were both assessed as binary variables (yes/no) to reflect modifiable health behaviors with known associations to headache-related symptom presentations and tinnitus. Body mass index (BMI) was calculated from measured height and weight, and treated as a continuous indicator of adiposity. Blood-based biomarkers included total cholesterol, high-density lipoprotein cholesterol (HDL-C), and serum uric acid, all analyzed as continuous variables due to their mechanistic relevance to vascular and metabolic health. Clinical diagnoses of hypertension, diabetes, cardiovascular disease (CVD), and cancer were included as binary variables, based on self-reported physician diagnoses, to account for comorbid conditions potentially influencing headache susceptibility in this population. This variable classification provided a theoretical foundation for stepwise modeling, facilitating the evaluation of how structural disadvantage and physiological burden jointly contribute to the likelihood of severe headache or migraine symptoms in peri- and postmenopausal women. Statistical analysis To account for the complex, multistage probability sampling design of NHANES and ensure nationally representative estimates, all analyses applied sampling weights, strata, and primary sampling units in accordance with NHANES analytic protocols. Design-adjusted descriptive analyses were performed to summarize baseline characteristics. Differences in categorical variables by severe headache or migraine symptom status were examined using the Rao–Scott corrected Chi-square test, while distributional differences in continuous variables were evaluated using the survey-weighted Mann–Whitney U test, given the presence of non-normality. The association between tinnitus constancy and self-reported severe headache or migraine symptoms was explored using a series of multivariable logistic regression models, with estimated effects presented as odds ratios (ORs) and corresponding 95% confidence intervals (CIs). A hierarchical modeling strategy was employed based on the Social Determinants of Health framework. Model 1 was unadjusted. Model 2 included structural determinants such as age, race/ethnicity, education level, marital status, and poverty income ratio. Model 3 incorporated additional intermediary determinants, including body mass index, total cholesterol, HDL-C, serum uric acid, smoking and drinking status, and binary indicators for hypertension, diabetes, cardiovascular disease, and cancer. Multicollinearity was assessed using variance inflation factors (VIF), and no evidence of multicollinearity was detected (all VIFs < 5). Model fit was evaluated using the Akaike Information Criterion (AIC), with lower values indicating better relative model performance. To further investigate variable importance and support feature selection, a multi-method ensemble approach was applied. First, variable importance was assessed using five distinct approaches: LASSO regression, logistic regression, random forest classification, the Boruta algorithm, and SHAP-based ranking from the LightGBM model. Predictors that consistently appeared among the top-ranked variables across these methods were considered highly influential. Subsequently, dependence plots were constructed for top-ranked predictors to visualize the nature and direction of their associations with headache-related symptom presentations. All analyses were conducted using R software (version 2024.12.1 + 563; Posit Software, PBC). A two-tailed P-value of less than 0.05 was considered statistically significant. Results Sample characteristics of the participants Table 1 presents the unweighted demographic characteristics of the study population. A total of 2,485 participants were included in the analysis, of whom 486 (19.56%) reported experiencing self-reported severe headache or migraine symptoms. Among those with these symptoms, 210 individuals (43.21%) also reported tinnitus symptoms. In contrast, among the 1,999 participants without such symptoms, 1,477 (73.89%) did not report tinnitus, while 522 (26.11%) did. These unweighted descriptive statistics provide an overview of the sample distribution across headache-symptom and tinnitus subgroups. Table 1Unweighted characteristics of the study populationCharacteristicTotalN = 2485Without self-reported severe headache or migraine symptomsN = 1999With self-reported severe headache or migraine symptomsN = 486Without tinnitusN = 1477With tinnitusN = 522Without tinnitusN = 276With tinnitusN = 210Age Mean (SD)64.3 (12.5)65.5 (12.1)66.8 (11.8)57.7 (12.4)58.1 (11.8) Median [Min, Max]64.0 [23.0, 85.0]65.0 [23.0, 85.0]67.0 [31.0, 85.0]56.0 [33.0, 85.0]58.0 [33.0, 85.0]Race Mexican American480 (19.3%)269 (18.2%)112 (21.5%)49 (17.8%)50 (23.8%) Non-Hispanic Black108 (4.3%)61 (4.1%)17 (3.3%)16 (5.8%)14 (6.7%) Non-Hispanic White1394 (56.1%)843 (57.1%)314 (60.2%)134 (48.6%)103 (49.0%) Other Hispanic424 (17.1%)264 (17.9%)57 (10.9%)69 (25.0%)34 (16.2%) Other race79 (3.2%)40 (2.7%)22 (4.2%)8 (2.9%)9 (4.3%)Education levels High school or below1491 (60.0%)851 (57.6%)330 (63.2%)175 (63.4%)153 (64.3%) Some college or higher994 (40.0%)626 (42.4%)192 (36.8%)101 (36.6%)75 (35.7%)Marital status Married/living with partner1288 (51.8%)761 (51.5%)256 (49.0%)148 (53.6%)123 (58.6%) Widowed/divorced/separated1087 (43.7%)652 (44.1%)244 (46.7%)113 (40.9%)78 (37.1%) Never married110 (4.4%)64 (4.3%)22 (4.2%)15 (5.4%)9 (4.3%)PIR Mean (SD)2.61 (1.59)2.73 (1.59)2.43 (1.58)2.49 (1.59)2.35 (1.58) Median [Min, Max]2.23 [0, 5.00]2.43 [0, 5.00]1.98 [0, 5.00]2.15 [0, 5.00]1.72 [0, 5.00]BMI (kg/m2) Mean (SD)29.1 (6.44)28.9 (6.34)29.1 (6.29)29.8 (6.67)29.5 (7.12) Median [Min, Max]28.1 [14.7, 62.5]28.0 [15.2, 62.5]28.0 [14.7, 52.1]29.0 [18.0, 59.9]28.6 [16.4, 57.2]Total cholesterol (mg/dL) Mean (SD)216 (40.0)216 (39.6)217 (41.7)216 (40.5)214 (37.7) Median [Min, Max]214 [93.0, 370]214 [93.0, 370]215 [122, 360]214 [124, 368]210 [124, 335]HDL-C (mg/dL) Mean (SD)58.4 (16.9)59.7 (17.0)56.4 (16.4)58.7 (17.1)54.7 (15.8) Median [Min, Max]56.0 [23.0, 154]57.0 [23.0, 142]54.0 [28.0, 114]56.0 [28.0, 154]52.5 [25.0, 111]Uric acid (mg/dL) Mean (SD)5.13 (1.41)5.14 (1.40)5.20 (1.41)5.13 (1.52)4.94 (1.33) Median [Min, Max]4.90 [0.40, 13.9]4.90 [1.40, 12.2]5.00 [2.20, 13.9]4.90 [1.60, 10.7]4.90 [0.40, 9.40]Smoking Yes1055 (42.5%)589 (39.9%)243 (46.6%)116 (42.0%)107 (51.0%) No1430 (57.5%)888 (60.1%)279 (53.4%)160 (58.0%)103 (49.0%)Drinking Yes1263 (50.8%)746 (50.5%)280 (53.6%)135 (48.9%)102 (48.6%) No1222 (49.2%)731 (49.5%)242 (46.4%)141 (51.1%)108 (51.4%)Hypertension Yes1247 (50.2%)715 (48.4%)294 (56.3%)139 (50.4%)99 (47.1%) No1238 (49.8%)762 (51.6%)228 (43.7%)137 (49.6%)111 (52.9%)Diabetes Yes385 (15.5%)213 (14.4%)98 (18.8%)44 (15.9%)30 (14.3%) No2100 (84.5%)1264 (85.6%)424 (81.2%)232 (84.1%)180 (85.7%)CVD Yes376 (15.1%)191 (12.9%)101 (19.3%)51 (18.5%)33 (15.7%) No2109 (84.9%)1286 (87.1%)421 (80.7%)225 (81.5%)177 (84.3%)Cancer Yes338 (13.6%)195 (13.2%)79 (15.1%)30 (10.9%)34 (16.2%) No2147 (85.4%)1282 (86.8%)443 (84.9%)246 (89.1%)176 (83.8%)Tinnitus constancy No tinnitus1753 (70.5%)1477 (100%)0 (0%)276 (100%)0 (0%) Occasional tinnitus384 (15.5%)0 (0%)272 (53.1%)0 (0%)112 (53.3%) Intermittent tinnitus180 (7.2%)0 (0%)120 (23.0%)0 (0%)60 (28.6%) Constant tinnitus166 (6.8%)0 (0%)130 (24.9%)0 (0%)38 (18.1%)PIR poverty-to-income ratio, BMI body mass index, HDL-C high-density lipoprotein cholesterol, CVD cardiovascular disease, unweighted descriptive counts; not used for inferential analyses Unweighted characteristics of the study population PIR poverty-to-income ratio, BMI body mass index, HDL-C high-density lipoprotein cholesterol, CVD cardiovascular disease, unweighted descriptive counts; not used for inferential analyses Table 2 presents the weighted characteristics of the study participants. The mean age of the sample was 60.66 years (SD = 12.51), and the average BMI was 28.81 (SD = 6.64). Participants with self-reported severe headache or migraine symptoms demonstrated a significantly different distribution of tinnitus constancy compared to those without such symptoms(P < 0.001). Among individuals without these symptoms, 73.88% reported no tinnitus, 13.01% occasional tinnitus, 5.54% intermittent tinnitus, and 7.57% constant tinnitus. In contrast, among those with self-reported severe headache or migraine symptoms, only 56.62% reported no tinnitus, while 23.48% experienced occasional tinnitus, 10.73% intermittent tinnitus, and 9.17% constant tinnitus. In addition to tinnitus, the presence of these headache-related symptoms was also associated with differences in several other characteristics. Statistically significant between-group differences were observed for age, educational attainment, PIR, BMI, HDL-C levels, and the prevalence of cardiovascular disease (all P < 0.05). Table 2Sample characteristics and comparison between participants with and without self-reported severe headache or migraine symptoms (weighted)CharacteristicN1OverallN = 31,666,7552Self-reported severe headache or migraine symptomsP-value3WithoutN = 24,656,9382WithN = 7,009,8182Age2,48560.66 ± (12.51)62.36 ± (12.29)54.71 ± (11.42) < 0.001Race2,4850.042 Mexican American480 (3.56%)381 (3.49%)99 (3.82%) Other Hispanic108 (5.40%)78 (4.74%)30 (7.72%) Non-Hispanic White1,394 (77.62%)1,157 (79.18%)237 (72.13%) Non-Hispanic Black424 (9.07%)321 (8.42%)103 (11.35%) Other race79 (4.35%)62 (4.17%)17 (4.98%)Education level2,4850.042 High school or below1,491 (51.53%)1,181 (49.80%)310 (57.59%) Some college or higher994 (48.47%)818 (50.20%)176 (42.41%)Marital Status2,4850.118 Married/living with partner1,288 (59.46%)1,017 (58.25%)271 (63.70%) Widowed/divorced/separated1,087 (36.47%)896 (37.80%)191 (31.77%) Never married110 (4.07%)86 (3.94%)24 (4.53%)PIR2,4852.96 ± (1.62)3.01 ± (1.61)2.78 ± (1.64)0.022BMI (kg/m2)2,48528.81 ± (6.64)28.61 ± (6.46)29.48 ± (7.19)0.035Total cholesterol (mg/dL)2,485216.53 ± (38.59)216.89 ± (38.78)215.26 ± (37.92)0.449HDL-C (mg/dL)2,48558.82 ± (17.03)59.24 ± (16.94)57.35 ± (17.30)0.041Uric acid (mg/dL)2,4855.04 ± (1.37)5.08 ± (1.37)4.92 ± (1.35)0.056Smoking2,4850.207 Yes1,055 (46.21%)832 (45.36%)223 (49.21%) No1,430 (53.79%)1,167 (54.64%)263 (50.79%)Drinking2,4850.499 Yes1,263 (55.84%)1,026 (56.29%)237 (54.25%) No1,222 (44.16%)973 (43.71%)249 (45.75%)Diabetes2,4850.345 Yes385 (11.70%)311 (11.36%)74 (12.88%) No2,100 (88.30%)1,688 (88.64%)412 (87.12%)Hypertension2,4850.210 Yes1,247 (44.84%)1,009 (45.62%)238 (42.08%) No1,238 (55.16%)990 (54.38%)248 (57.92%)CVD2,4850.049 Yes376 (13.87%)292 (13.02%)84 (16.86%) No2,109 (86.13%)1,707 (86.98%)402 (83.14%)Cancer2,4850.647 Yes338 (14.43%)274 (14.25%)64 (15.08%) No2,147 (85.57%)1,725 (85.75%)422 (84.92%)Tinnitus2,485 < 0.001 No tinnitus1,753 (70.06%)1,477 (73.88%)276 (56.62%) Occasional tinnitus384 (15.33%)272 (13.01%)112 (23.48%) Intermittent tinnitus180 (6.69%)120 (5.54%)60 (10.73%) Constant tinnitus168 (7.92%)130 (7.57%)38 (9.17%)1N not missing (unweighted)2Mean ± (SD); n (unweighted) (%)3Adjustment Chi-squared test with Rao and Scott's second-order correction; Mann–Whitney test for complex survey samples Sample characteristics and comparison between participants with and without self-reported severe headache or migraine symptoms (weighted) 1N not missing (unweighted) 2Mean ± (SD); n (unweighted) (%) 3Adjustment Chi-squared test with Rao and Scott's second-order correction; Mann–Whitney test for complex survey samples Association between tinnitus constancy and self-reported severe headache or migraine symptoms among peri- and postmenopausal women As shown in Table 3, a consistent association was observed between the constancy of tinnitus and the odds of self-reported severe headache or migraine symptoms among peri- and postmenopausal women. Table 3Association between tinnitus constancy and self-reported severe headache or migraine symptoms among peri- and postmenopausal women (weighted)ExposureModel 1Model 2Model 3OR (95% CI)P valueOR (95% CI)P valueOR (95% CI)P valueTinnitus constancy No tinnitus1 (Ref.)1 (Ref.)1 (Ref.) Occasional tinnitus2.35 (1.54,3.60) < 0.0012.23 (1.43,3.48) < 0.0012.29 (1.44,3.63)0.001 Intermittent tinnitus2.53(1.61, 3.97) < 0.0012.57 (1.57,4.21) < 0.0012.62 (1.57,4.35) < 0.001 Constant tinnitus1.58(1.10, 2.28)0.0161.96 (1.31,2.95)0.0021.93 (1.26,2.95)0.004 P for trend0.009 < 0.0010.002 Age0.94 (0.93,0.96) < 0.0010.94 (0.92,0.95) < 0.001Race Mexican American1 (Ref.)1 (Ref.) Other Hispanic1.63 (0.88,3.04)0.1171.63 (0.84,3.15)0.138 Non-Hispanic White1.24 (0.83,1.84)0.2801.17 (0.79,1.73)0.413 Non-Hispanic Black1.54 (0.97,2.43)0.0651.37 (0.87,2.15)0.162Other race1.37 (0.56,3.34)0.4751.24 (0.50,3.07)0.633Education level High school or below1 (Ref.)1 (Ref.) Some colleges or higher0.70 (0.50,0.97)0.0340.73 (0.52,1.02)0.064Marital status Married/living with partner1 (Ref.)1 (Ref.) Widowed/divorced/separated0.87 (0.67,1.13)0.2780.88 (0.67,1.16)0.349 Never married0.85 (0.44,1.66)0.6260.93 (0.49,1.82)0.821 PIR0.90 (0.83,0.99)0.0260.92 (0.84,1.01)0.065 BMI (kg/m2)1.00 (0.99,1.02)0.666Total cholesterol (mg/dL)1.00 (1.00,1.00)0.828 HDL-C (mg/dL)1.01 (1.00,1.01)0.122 Uric acid (mg/dL)0.99 (0.88,1.11)0.905Smoking Yes1 (Ref.) No1.08 (0.81,1.44)0.585Drinking Yes1 (Ref.) No1.33 (1.04,1.71)0.028Hypertension Yes1 (Ref.) No0.97 (0.74,1.29)0.846Diabetes Yes1 (Ref.) No0.96 (0.67,1.39)0.842CVD No1 (Ref.) Yes2.01 (1.35,2.99)0.001Cancer Yes1 (Ref.) No0.74 (0.54,1.03)0.075Model 1: not adjustedModel 2: adjusted for predisposing and enabling factors (age, race, education level, marital status, and PIR)Model 3: additionally adjusted for need factors (BMI, TC, HDL-C, Uric Acid, smoking, drinking, hypertension, diabetes, CVD, and cancer)PIR poverty-to-income ratio, BMI body mass index, TC total cholesterol, HDL-C high-density lipoprotein cholesterol, CVD cardiovascular disease, OR odds ratio, CI confidence interval Association between tinnitus constancy and self-reported severe headache or migraine symptoms among peri- and postmenopausal women (weighted) Model 1: not adjusted Model 2: adjusted for predisposing and enabling factors (age, race, education level, marital status, and PIR) Model 3: additionally adjusted for need factors (BMI, TC, HDL-C, Uric Acid, smoking, drinking, hypertension, diabetes, CVD, and cancer) PIR poverty-to-income ratio, BMI body mass index, TC total cholesterol, HDL-C high-density lipoprotein cholesterol, CVD cardiovascular disease, OR odds ratio, CI confidence interval Across all three models, we observed a statistically significant positive trend across tinnitus constancy categories (P for trend < 0.05), indicating that the distribution of symptom severity categories was associated with differing odds of self-reported severe headache or migraine symptoms. Model 3 demonstrated the best model fit, as evidenced by the lowest Akaike Information Criterion (AIC = 2389.74), compared to Model 1 (AIC = 2574.37) and Model 2 (AIC = 2396.77). After adjusting for potential confounders, the fully adjusted Model 3 revealed that, relative to women without tinnitus, those with occasional tinnitus had 2.29 times higher odds of self-reported severe headache or migraine symptoms (95% CI 1.44–3.63, P = 0.001). Women with intermittent tinnitus had 2.62 times higher odds (95% CI 1.57–4.35, P < 0.001), while those with constant tinnitus also showed significantly elevated odds (OR = 1.93, 95% CI 1.26–2.95, P = 0.004). Importance of variables To identify the key determinants of self-reported severe headache or migraine symptoms among peri- and postmenopausal women, we evaluated variable importance using five complementary approaches based on baseline data: LASSO regression, logistic regression, random forest classification, the Boruta algorithm, and a SHAP-based LightGBM model. As shown in Fig. 1A–C, all three methods—LASSO regression, logistic regression, and random forest—consistently highlighted age as the most influential factor, followed by tinnitus constancy. Similarly, results from the Boruta algorithm (Fig. 1D–E) confirmed that age and tinnitus constancy were the top two predictive features. The SHAP-based LightGBM model yielded consistent findings, identifying these two variables as the primary determinants of self-reported severe headache or migraine symptoms in this population. For improved clarity, high-resolution versions of each individual subpanel are also provided in the Supplementary Materials (Figures S1–S7). Importance of variables identified by five complementary algorithms. A LASSO regression showing selected predictors and their coefficients. B Logistic regression–based variable importance. C Random forest classification ranking of feature importance. D Boruta algorithm displaying the importance history of variables across iterations. E Final confirmed important features based on Boruta selection. F SHAP beeswarm plot visualizing the distribution and magnitude of feature contributions across individual predictions. G SHAP bar plot illustrating the global mean absolute SHAP values for each variable. High-resolution standalone versions of all subpanels are available in the Supplementary Materials (Figures S1–S7) Dependence visualization To further examine the effect patterns of the influential features, SHAP dependence plots were constructed for variables that consistently demonstrated high importance across all selection methods (Fig. 2). The plots confirmed that age and tinnitus constancy were the two strongest contributors to the likelihood of reporting severe headache or migraine symptoms among peri- and postmenopausal women. Specifically, younger age and higher levels of tinnitus constancy were associated with increased odds of self-reported severe headache or migraine symptoms. SHAP dependence plots for the top two most important features. A Age and its marginal effect on predicted migraine risk. B Tinnitus constancy and its graded association with migraine probability. C Age and predicted migraine risk, with data points color-encoded by tinnitus constancy to highlight potential interaction effects. D Tinnitus constancy and predicted migraine probability, with data points color-encoded by age to reveal cross-feature modulation. Tinnitus constancy was coded as 0 = no tinnitus, 1 = occasional tinnitus, 2 = intermittent tinnitus, and 3 = constant tinnitus. SHAP values were derived from the LightGBM model Discussion This study identified a significant association between tinnitus constancy and the odds of self-reported severe headache or migraine symptoms among peri- and postmenopausal women. Across regression models and machine-learning approaches, intermittent tinnitus consistently demonstrated the strongest association, whereas age was a robust inverse predictor. The convergence of results across multiple analytical frameworks indicates that temporal auditory patterns—rather than the mere presence of tinnitus—may meaningfully relate to headache-related symptom presentation in peri- and postmenopausal women. The observed association between tinnitus and migraine is consistent with prior reports indicating an elevated prevalence of tinnitus symptoms among individuals with migraine and suggesting shared pathophysiological mechanisms, including central sensitization and sensory network dysfunction [21, 22]. However, most previous studies did not stratify by menopausal status or consider the temporal pattern of tinnitus as a clinically relevant dimension. By incorporating a refined classification of tinnitus temporal patterns, the present study shows that the odds of migraine differ across tinnitus constancy categories, with intermittent tinnitus demonstrating the strongest association in peri- and postmenopausal women. This study contributes to existing research by moving beyond a binary definition of tinnitus and describing how tinnitus constancy may serve as a useful descriptive characteristic in epidemiologic analyses. Capturing the degree and variability of tinnitus exposure may provide a more accurate reflection of the cumulative neurosensory burden experienced by migraine-prone individuals. These findings underscore the need to integrate tinnitus profiling into future studies of migraine subtypes and highlight its relevance for research involving peri- and postmenopausal populations. Interpretation of model differences is also essential. While odds ratios varied slightly across sequentially adjusted models, the direction and pattern remained stable, and the fully adjusted model offered the most interpretable estimates. Additional variables—namely drinking status and cardiovascular disease—also reached significance, consistent with population-based studies showing co-occurrence of lifestyle and cardiometabolic factors with headache disorders in midlife adults [35, 36]. These were covariates rather than exposures of primary interest; their significance reflects background comorbidity rather than unique mechanistic pathways. Multiple neurobiological and endocrine mechanisms have been proposed in previous studies as potential contributors to this non-monotonic association. First, intermittent tinnitus, due to its fluctuating and unpredictable nature, may elicit heightened emotional and physiological responses, disrupting sensory filtering and triggering increased vigilance [24, 37]. Such responses have been suggested to involve activation of limbic circuits and thalamocortical pathways [38, 39], both of which have been implicated in central sensitization, a key process in migraine pathophysiology [40]. Second, fluctuating auditory input may interfere with habituation processes [41], heightening cortical excitability and attentional demand [42], and predisposing migraine-prone individuals to attacks [43, 44]. Third, serotonergic dysregulation, a shared feature of both tinnitus and migraine, may be more pronounced in intermittent symptoms, contributing to dysregulated pain and auditory processing [15, 18, 45]. Importantly, these mechanisms are likely modulated by hormonal changes during the menopausal transition, which is characterized by declining levels of estrogen and progesterone [46]. Estrogen has known neuromodulatory effects, including the regulation of serotonin synthesis, vascular tone, and cortical excitability [47]. Declines in estrogen levels have been associated with increased migraine frequency [48, 49], reduced pain thresholds [50], and altered auditory perception [51]. Moreover, estrogen withdrawal may impair GABAergic inhibition, further amplifying neuronal excitability and emotional reactivity in response to internal stimuli such as tinnitus [52, 53]. These hormone-related changes may increase vulnerability to both conditions. However, these mechanistic pathways are derived from prior experimental and clinical work and cannot be directly tested in the present analysis, which lacks physiological or neuroendocrine measurements; they should therefore be regarded as hypothesis-generating rather than definitive explanations of the observed associations. Beyond these hypotheses, clinical literature indicates that tinnitus may appear as part of the prodromal or aura phases of migraine, manifesting as transient auditory disturbances preceding headache onset [18, 54]. Given the cross-sectional nature of NHANES, the temporal sequence between tinnitus and headache-related symptoms cannot be determined, and this alternative explanation offers a plausible non-causal interpretation of the observed association. In contrast, constant tinnitus, although persistent, may permit partial neural adaptation over time [24]. The steady presence of auditory stimuli may reduce novelty, limit emotional salience, and allow for habituation within auditory and affective neural circuits [55]. Neuroimaging studies have suggested that chronic tinnitus may be associated with reduced activation in the default mode network (DMN) and stabilization of auditory cortex activity, potentially buffering against migraine triggers [56, 57]. Age consistently emerged as the strongest inverse predictor across regression and machine-learning analyses. This pattern is consistent with epidemiological findings that migraine prevalence peaks in early adulthood and declines with age [57–63], and that symptoms may improve after menopause due to stabilization of estrogen levels [9, 64]. However, due to the absence of hormone measurements in these survey cycles, the present study cannot disentangle age-related from menopause-related influences; therefore, age is interpreted as reflecting broader life-course patterns rather than direct menopausal effects. From a clinical perspective, these findings reinforce the importance of incorporating structured tinnitus assessment when evaluating headache-related symptoms in peri- and postmenopausal women. Younger age and intermittent tinnitus—identified consistently as top predictors—may indicate increased central sensitization or reduced sensory habituation, potentially aiding in early identification of headache-related risk phenotypes. Integrating sensory symptom profiling into midlife women’s health evaluations, including in primary care and gynecologic settings, may facilitate more precise screening and management strategies. This study has several limitations. First, its cross-sectional design prevents causal inference. Second, NHANES variable constraints limited clinical specificity. Critically, the survey item “severe headache or migraine” does not differentiate migraine from cluster headache, tension-type headache, or secondary headache disorders and therefore does not constitute a clinically validated diagnosis. Accordingly, all analyses were interpreted in terms of self-reported severe headache or migraine symptoms rather than definitive migraine. Third, menopausal status was inferred from self-reported menstrual history. Fourth, the tinnitus data lacked detail on loudness, chronicity, laterality, and treatment history. Fifth, psychiatric and sleep variables were unavailable for this age group, precluding adjustment for depression, anxiety, or sleep disturbances—conditions known to affect both tinnitus and migraine [65]. Sixth, the analysis was necessarily limited to the 1999–2004 NHANES cycles, the most recent cycles containing relevant headache questions. Finally, the study population was limited to peri- and postmenopausal women, and results may not generalize to younger adults. Despite these limitations, the consistency of findings across multiple analytic approaches suggests that tinnitus constancy—particularly intermittent patterns—may be a meaningful sensory correlate of headache-related symptoms in midlife women. Future longitudinal studies incorporating detailed auditory, neuroendocrine, and prodromal symptom data are needed to clarify these relationships. Conclusions This study identified a significant and consistent association between tinnitus constancy and the odds of self-reported severe headache or migraine symptoms among peri- and postmenopausal women, with intermittent tinnitus showing the highest odds ratio within this pattern. Age was inversely associated with these symptoms and emerged as the most influential predictor across multiple analytical approaches. These findings suggest that both the temporal pattern of tinnitus and age-related factors contribute meaningfully to headache-related symptom vulnerability in peri- and postmenopausal women. The results underscore the potential utility of incorporating sensory symptom profiling into frameworks for headache and migraine-related symptom assessment and phenotyping, without implying predictive or causal inference. Further longitudinal and mechanistic research is warranted to clarify causal pathways and inform targeted strategies for early detection and intervention in this clinically vulnerable population. Supplementary Information Supplementary file 1 Acknowledgements Not applicable. Author contributions CY was responsible for the conceptualization of the study, methodological design, statistical analysis, interpretation of results, drafting of the original manuscript, and critical revision for intellectual content. JL contributed substantially to the revision process, including refinement of analytical descriptions, rewriting and restructuring of the Discussion, responding to reviewer comments, and improving the clarity and accuracy of the manuscript. DHM provided critical revisions to the manuscript. All authors reviewed and approved the final version of the manuscript. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability The data underlying this study are freely accessible to all interested researchers through the National Health and Nutrition Examination Survey (NHANES) database. The data can be obtained by visiting the following website: (https://www.cdc.gov/nchs/nhanes/?CDC_AAref_Val=https:/www.cdc.gov/nchs/nhanes/index.htm). Declarations Ethics approval and consent to participate The NHANES protocols were approved by the Ethics Review Board of the National Center for Health Statistics, Centers for Disease Control and Prevention. All participants provided written informed consent prior to the home interview and physical examination procedures. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary file 1 Data Availability Statement The data underlying this study are freely accessible to all interested researchers through the National Health and Nutrition Examination Survey (NHANES) database. The data can be obtained by visiting the following website: (https://www.cdc.gov/nchs/nhanes/?CDC_AAref_Val=https:/www.cdc.gov/nchs/nhanes/index.htm).