Abstract
Diabetic retinopathy (DR) significantly impairs patients’ psychological well-being and quality of life, yet the association between DR and suicidal ideation (SI) remains unclear. This study investigates this potential association and its consistency across various population subgroups. We performed a cross-sectional analysis using data from 7 cycles (2005–2006 through 2017–2018) of the National Health and Nutrition Examination Survey. The final analytical sample comprised 4006 adults with diabetes. Weighted multivariate regression analysis was used to assess the association between DR and SI. Subgroup analyses, interaction analyses, and a series of sensitivity analyses were used to explore the stability of the results. After sequential adjustment for demographic characteristics, lifestyle factors, comorbidities, and depressive symptoms, DR was associated with an increased risk for SI (odds ratio = 1.70, 95% confidence interval: 1.08–2.68, P = .023). Subgroup analyses showed generally consistent positive associations across most subgroups, and no significant interaction remained after correction for multiple comparisons. Sensitivity analyses confirmed robustness (multiple imputation and diabetes-related adjustment remained significant). In a sensitivity analysis using a fundus photography-based DR definition (limited to 2005–2008 cycles with a small sample size and few SI events), the estimate was nonsignificant but directionally consistent (odds ratio = 1.27; P = .607). DR is associated with increased SI among United States diabetic adults, independent of depressive symptoms. Given the cross-sectional design, these findings should be interpreted cautiously and confirmed in future prospective studies.
Keywords: cross-sectional study, diabetic retinopathy (DR), NHANES, suicidal ideation (SI)
1. Introduction
Suicide, defined as a deliberate act of self-harm resulting in death, represents a major global public health challenge.[1] Global statistics indicate that approximately 746,000 individuals died by suicide in 2021, corresponding to a rate of 9.0 per 100,000 people.[2] Suicidal ideation (SI) refers to thoughts about ending one’s own life or self-harm,[3] and it is an important precursor to suicidal behavior and a key predictor of suicide risk.[4] In clinical practice, heightened attention should be paid to screening for SI, actively exploring its related risk factors, and developing targeted intervention strategies.
A substantial body of research has established significant links between SI and various factors, including mood disorders, chronic medical conditions, and diminished quality of life.[5–7] Recent evidence suggests that visual impairment is associated with a higher risk of SI, although the strength of this association varies across study populations and outcome definitions.[8,9] Another meta-analysis showed that diabetes is associated with suicide-related outcomes, including SI, suicide attempts, and death by suicide.[10] Compared with general visual impairment, diabetic retinopathy (DR) is more diabetes-specific, because visual impairment may also result from various non-diabetic eye diseases. Compared with other diabetic complications, DR has the dual characteristics of a chronic diabetic complication and a direct threat to visual function. It may affect mental health through visual decline, fear of blindness, treatment burden, and reduced quality of life.
However, direct evidence on the relationship between DR and SI remains limited. A previous hospital-based study of older adults with visual impairment reported a positive association between DR and SI.[11] However, that study was based mainly on older adults from a single center in Thailand and did not sufficiently adjust for key psychological factors such as depressive symptoms, which limits the generalizability of its findings. Therefore, among adults with diabetes, especially after adjustment for important confounders such as depressive symptoms, population-based evidence remains lacking as to whether DR is associated with SI.
Therefore, this study utilized data from the nationally representative National Health and Nutrition Examination Survey (NHANES) to investigate the relationship between DR and SI and to assess whether this relationship varies among demographic and clinical subgroups.
2. Methods
2.1. Study design and data source
This study constituted a cross-sectional analysis of data collected from 7 cycles (2005–2018) of the NHANES (http://wwwn.cdc.gov/nchs/nhanes). NHANES is conducted by the National Center for Health Statistics (NCHS) to evaluate the health and nutritional status of the civilian, noninstitutionalized United States (U.S.) population. The NCHS Research Ethics Review Board approved the NHANES protocol, and all participants provided written informed consent. The analysis of de-identified, publicly available NHANES data was exempt from further institutional review.
Our analysis incorporated 7 NHANES cycles (2005–2018), focusing on adults (≥ 20 years) with diabetes (n = 7353). Participants were excluded if they had incomplete depression screening questionnaire data (n = 919) or missing DR data (n = 2025), leaving 4409 participants. We further excluded participants with missing covariate data, including marital status (n = 4), education level (n = 5), body mass index (BMI, n = 89), smoking status (n = 1), alcohol use (n = 260), hypertension (n = 1), congestive heart failure (n = 22), or angina pectoris (n = 21). A total of 4006 participants were ultimately included in the final statistical analysis. Figure 1 presents a detailed flowchart outlining the data inclusion and exclusion procedures.
Figure 1.
Flow diagram of the sample selection from the NHANES 2005 to 2018. n = number of participants, NHANES = National Health and Nutrition Examination Survey, PHQ = Patient Health Questionnaire.
2.2. Diagnostic criteria for diabetes and DR
Diabetes mellitus was diagnosed when participants met at least one of these criteria: hemoglobin A1c (HbA1c) levels ≥ 6.5%; fasting plasma glucose ≥ 7.0 mmol/L; random plasma glucose or 2-hour plasma glucose during an oral glucose tolerance test ≥ 11.1 mmol/L, use of glucose-lowering medication/insulin, or self-reported physician-confirmed diabetes diagnosis.[12,13]
DR status was determined using a standardized item from the Diabetes Interview questionnaire: “Has a physician ever informed you that diabetes has affected your eyes or that you have retinopathy?” Affirmative responses were classified as positive for DR.[14–16]
2.3. Evaluation of depressive symptoms and SI
Mental health assessment was conducted using the validated 9-item Patient Health Questionnaire (PHQ).[17] This instrument’s structure corresponds to the diagnostic and statistical manual of mental disorders, fifth edition diagnostic criteria for major depressive disorder. Participants rated symptom frequency over the preceding 2 weeks using a 4-point scale (0 = never; 1 = several days; 2 = more than half the days; 3 = nearly daily), with total scores ranging from 0 to 27.[18,19]
SI was specifically assessed using item 9 (DPQ090) of the PHQ-9, which evaluates thoughts of death or self-harm. Responses indicating occurrence “several days,” “more than half the days,” or “nearly every day” were classified as positive for SI, while “never” responses indicated absence of such ideation.[20–22]
The PHQ-8 score (sum of items 1–8, excluding the suicide item) was used to assess depressive symptoms. Scores ≥ 10 indicated clinically significant depression, while scores < 10 suggested absence of depressive symptoms.[23–34]
2.4. Covariates
Based on evidence from prior literature and clinical practice,[25–37] we adjusted for the following potential confounders: Sociodemographic and lifestyle factors included gender, age, marital status, race/ethnicity, BMI, physical activity level, educational attainment, smoking history, and alcohol consumption. Physical activity was quantified using metabolic equivalent task (MET) values and calculated according to the type, frequency, and duration of weekly physical activity as follows: physical activity level = MET value of the activity × weekly frequency × duration of each activity session. Different activities were assigned different MET values according to the NHANES-recommended MET values for specific physical activities.[28,29] Alcohol consumption was classified into 5 categories: never drinking, former drinking, mild drinking, moderate drinking, and heavy drinking. Never drinking was defined as never having consumed alcohol, whereas former drinking was defined as having consumed alcohol previously but not drinking during the past year. Heavy drinking was defined as ≥ 3 drinks/day for women, ≥ 4 drinks/day for men, or binge drinking on ≥ 5 days per month; binge drinking was defined as ≥ 4 drinks for women or ≥ 5 drinks for men on one occasion. Moderate drinking was defined as ≥ 2 drinks/day for women, ≥ 3 drinks/day for men, or binge drinking on ≥ 2 days per month. Mild drinking was defined as alcohol consumption that did not meet the criteria for moderate or heavy drinking.[30,31] Smoking status was classified as never, former, or current smoking. Never smokers were defined as participants who had never smoked or had smoked fewer than 100 cigarettes in their lifetime. Former smokers were defined as those who had smoked at least 100 cigarettes in their lifetime but had quit smoking, whereas current smokers were defined as those who had smoked at least 100 cigarettes in their lifetime and were still smoking. Educational attainment was categorized as college graduate or above, some college or associate degree, high school graduate/GED or equivalent, grades 9 to 11, and < 9th grade. Comorbid conditions included hypertension, congestive heart failure, angina pectoris, and hyperlipidemia. Depressive symptoms, assessed using the PHQ-8 as described above, were also included as covariates in the analysis.
2.5. Statistical analysis
All regression analyses were conducted using survey-weighted procedures to account for the complex multistage sampling design of NHANES. The 2-year Mobile Examination Center examination sample weights were applied, and because 7 survey cycles (2005–2018) were combined, the weights were divided by 7 according to the recommendations of the NCHS. The masked variance pseudo-stratum and masked variance pseudo-primary sampling unit variables were incorporated to obtain nationally representative estimates and appropriate standard errors.
Continuous variables were compared using Student’s t-tests or Wilcoxon rank-sum tests, as appropriate, and are reported as mean ± standard deviation. Categorical variables were compared using χ2 tests and are reported as frequencies (percentages). Given the potential for selection bias related to the exclusion of participants with incomplete depressive symptom questionnaire data, we also performed a comprehensive comparison of baseline characteristics between the included participants (n = 6434) and those excluded from the analysis (n = 919). Weighted multivariate regression models were used to analyze the association between DR and SI. Model I: No adjustment for covariates. Model II: Adjusted for demographic variables (gender, age, race/ethnicity, marital status, education level). Model III: Further adjusted for lifestyle variables (BMI, MET, smoking, alcohol use) in addition to Model II. Model IV: Further adjusted for comorbidities (hypertension, congestive heart failure, angina pectoris, hyperlipidemia) in addition to Model III. Model V: Further adjusted for depressive status (based on PHQ-8 score) in addition to Model IV.
To evaluate the consistency of the association between DR and SI, we conducted subgroup analyses and interaction analyses based on gender, age group (< 65 years vs 65 years or older), and comorbid conditions (hypertension, hyperlipidemia, angina, congestive heart failure). To reduce the risk of false-positive findings from multiple testing, P values for interaction terms across all subgroup analyses were further adjusted using the Holm, Hochberg, Hommel, Bonferroni, Benjamini–Hochberg, Benjamini–Yekutieli, and false discovery rate methods.
Three sensitivity analyses were performed to ensure the robustness of the findings. First, to control for potential bias caused by missing covariate data, we performed multiple imputation for all covariates with missing values using chained equations, generating 5 imputed datasets. We then repeated the multivariable logistic regression models used in the main analysis. Second, to further assess the stability of the observed association, diabetes duration, glycated hemoglobin (HbA1c), and insulin use were additionally included as covariates in the multivariable regression models. Third, considering that the definition of DR based on patients’ self-reported information may be subject to non-specificity and underdiagnosis, we performed a sensitivity analysis using a fundus examination-based definition of DR. Specifically, with reference to recently published studies, DR status was redefined using retinopathy severity grading results derived from fundus photography in the NHANES ophthalmic examination module. During the NHANES 2005 to 2008 cycles, participants underwent 45° non-mydriatic digital fundus photography. Fundus images were evaluated by trained graders according to the NHANES Digital Grading Protocol and the early treatment DR study-related grading criteria. The severity grade of the worse eye was used to determine overall DR severity. Based on these grading results, DR severity was classified into 4 categories: no DR, mild non-proliferative DR, moderate/severe non-proliferative DR, and proliferative DR.[32,33] Because this objective definition was available only for the 2005 to 2008 cycles, and the number of SI events was limited within each DR severity subgroup, stable regression analyses across multiple DR severity categories could not be performed. Therefore, we further collapsed the exposure variable into a binary variable (DR vs no DR) and repeated the analysis to assess the robustness of the findings.
Statistical analyses were performed using R 4.2.2 (http://www.R-project.org, The R Foundation) and Free Statistics software version 2.1.1 (Clinical Scientists). A 2-sided P value < .05 was considered statistically significant.
3. Results
3.1. Baseline characteristics
Baseline characteristics of the participants are shown in Table 1. A total of 4006 participants, weighted to represent a population of 18,128,673, were included in the study. Among them, 221 (weighted to 860,183) had SI, with an unweighted prevalence of 5.5% (weighted 4.7%). When stratified by SI status, diabetic patients with SI were more likely to be female; of Mexican American or other Hispanic ethnicities; unmarried, including divorced, separated, widowed, or never married; to have lower educational attainment levels; higher BMI; a history of smoking or former alcohol use; and comorbidities such as congestive heart failure and depression. The proportion of participants with DR was higher among those with SI than among those without SI (28.1% vs 20.3%, P = .005), suggesting a potential association between DR and SI.
Table 1.
Baseline characteristics of participants.
| Variables | Total | Without SI | With SI | P value |
|---|---|---|---|---|
| n = 4006 | n = 3785 (94.5%) | n = 221 (5.5%) | ||
| N (Weight) | 18,128,673 | 17,268,490 (95.3%) | 860,183 (4.7%) | |
| Age (years, Mean ± SD) | 61.3 ± 12.9 | 61.4 ± 12.9 | 59.1 ± 12.7 | .010 |
| Gender, n (%) | .009 | |||
| Male | 2047 (51.1) | 1953 (51.6) | 94 (42.5) | |
| Female | 1959 (48.9) | 1832 (48.4) | 127 (57.5) | |
| Race Ethnicity, n (%) | .026 | |||
| Mexican American | 716 (17.9) | 669 (17.7) | 47 (21.3) | |
| Other Hispanic | 410 (10.2) | 376 (9.9) | 34 (15.4) | |
| Non-Hispanic White | 1400 (34.9) | 1326 (35) | 74 (33.5) | |
| Non-Hispanic Black | 1108 (27.7) | 1057 (27.9) | 51 (23.1) | |
| Other Race-including multiracial | 372 (9.3) | 357 (9.4) | 15 (6.8) | |
| Marital Status, n (%) | .008 | |||
| Married | 2226 (55.6) | 2125 (56.1) | 101 (45.7) | |
| Widowed | 565 (14.1) | 533 (14.1) | 32 (14.5) | |
| Divorced | 541 (13.5) | 509 (13.4) | 32 (14.5) | |
| Separated | 160 (4.0) | 148 (3.9) | 12 (5.4) | |
| Never married | 358 (8.9) | 325 (8.6) | 33 (14.9) | |
| Living with partner | 156 (3.9) | 145 (3.8) | 11 (5) | |
| Education Level, n (%) | < .001 | |||
| < 9th Grade | 686 (17.1) | 632 (16.7) | 54 (24.4) | |
| 9–11th Grade | 674 (16.8) | 626 (16.5) | 48 (21.7) | |
| High School | 906 (22.6) | 864 (22.8) | 42 (19) | |
| Some College or AA degree | 1115 (27.8) | 1053 (27.8) | 62 (28.1) | |
| College Graduate or above | 625 (15.6) | 610 (16.1) | 15 (6.8) | |
| BMI (kg/m2, Mean ± SD) | 32.5 ± 7.6 | 32.5 ± 7.5 | 33.9 ± 8.0 | .007 |
| Physical activity (MET-min/wk, Mean ± SD) |
1937.6 ± 4372.2 | 1953.6 ± 4383.6 | 1663.4 ± 4172.3 | .338 |
| Smoking, n (%) | < .001 | |||
| Never | 2019 (50.4) | 1933 (51.1) | 86 (38.9) | |
| Former | 1363 (34.0) | 1284 (33.9) | 79 (35.7) | |
| Current | 624 (15.6) | 568 (15.0) | 56 (25.3) | |
| Alcohol, n (%) | < .001 | |||
| Never | 741 (18.5) | 707 (18.7) | 34 (15.4) | |
| Former | 1116 (27.9) | 1026 (27.1) | 90 (40.7) | |
| Mild | 1293 (32.3) | 1242 (32.8) | 51 (23.1) | |
| Moderate | 372 (9.3) | 354 (9.4) | 18 (8.1) | |
| Heavy | 484 (12.1) | 456 (12.0) | 28 (12.7) | |
| Hypertension, n (%) | .659 | |||
| No | 1011 (25.2) | 958 (25.3) | 53 (24.0) | |
| Yes | 2995 (74.8) | 2827 (74.7) | 168 (76.0) | |
| Congestive heart failure, n (%) | < .001 | |||
| No | 3605 (90.0) | 3423 (90.4) | 182 (82.4) | |
| Yes | 401 (10.0) | 362 (9.6) | 39 (17.6) | |
| Angina pectoris, n (%) | .113 | |||
| No | 3715 (92.7) | 3516 (92.9) | 199 (90.0) | |
| Yes | 291 (7.3) | 269 (7.1) | 22 (10.0) | |
| Hyperlipidemia, n (%) | .773 | |||
| No | 554 (13.8) | 522 (13.8) | 32 (14.5) | |
| Yes | 3452 (86.2) | 3263 (86.2) | 189 (85.5) | |
| Depression, n (%) | < .001 | |||
| No | 3490 (87.1) | 3412 (90.1) | 78 (35.3) | |
| Yes | 516 (12.9) | 373 (9.9) | 143 (64.7) | |
| Retinopathy, n (%) | .005 | |||
| No | 3177 (79.3) | 3018 (79.7) | 159 (71.9) | |
| Yes | 829 (20.7) | 767 (20.3) | 62 (28.1) |
AA = associate degree, BMI = body mass index, MET = metabolic equivalent task, n = number of participants, SD = standard deviation, SI = suicidal ideation.
As shown in Table 1, Supplemental Digital Content 1, the comparison of baseline characteristics between the included and excluded populations showed significant differences in several key characteristics (P < .05). Compared with the included group, the excluded group was older, had a lower proportion of non-Hispanic White participants, a higher proportion of other races, lower educational attainment, a higher proportion of widowed participants, lower BMI, lower MET values, a very high proportion of missing alcohol-use information, a higher prevalence of congestive heart failure, and a lower reported prevalence of hyperlipidemia. These characteristics are often associated with poorer mental health status and a higher risk of suicide-related outcomes. Notably, the prevalence of DR was slightly higher in the excluded group (17.0% vs 14.5%, P = .036).
3.2. Association between DR and SI
The results of the survey-weighted logistic regression analyses are presented in Table 2. Across all 5 models, DR was significantly and positively associated with SI. In Model I, without adjustment for covariates, DR was associated with a 90% higher odds of SI (odds ratio [OR] = 1.90, 95% confidence interval [CI]: 1.28–2.84, P = .002). After sequential adjustment for demographic characteristics (Model II), lifestyle factors (Model III), and comorbidities (Model IV), the magnitude of the association was slightly attenuated but remained statistically significant. Even after further adjustment for depressive status in Model V, the association between DR and SI remained significant (OR = 1.70, 95% CI: 1.08–2.68, P = .023), suggesting that DR remained associated with SI after adjustment for depressive symptoms.
Table 2.
Multivariate regression analysis of the association between DR and SI.
| Variable | Without DR OR (95% CI) |
With DR OR (95% CI) |
P value |
|---|---|---|---|
| Model Ⅰ | 1.00 (Ref) | 1.90 (1.28–2.84) | .002 |
| Model Ⅱ | 1.00 (Ref) | 1.83 (1.19–2.82) | .006 |
| Model Ⅲ | 1.00 (Ref) | 1.81 (1.17–2.80) | .009 |
| Model Ⅳ | 1.00 (Ref) | 1.73 (1.09–2.73) | .020 |
| Model Ⅴ | 1.00 (Ref) | 1.70 (1.08–2.68) | .023 |
Model I: unadjusted. Model II: adjusted for Age + Gender + Race Ethnicity + Marital Status + Education Level. Model III: Model II + BMI + MET + Smoking + Alcohol. Model IV: Model III + Hypertension + Congestive heart failure + Angina pectoris + Hyperlipidemia. Model V: Model IV + Depression.
BMI = body mass index, CI = confidence interval, DR = diabetic retinopathy, MET = metabolic equivalent task, OR = odds ratio, SI = suicidal ideation.
Subgroup analyses are shown in Table 3. In most subgroups, DR was positively associated with SI, with ORs >1, which was generally consistent with the findings from the multivariable logistic regression models presented in Table 2. A statistically significant interaction was observed only for angina status (P for interaction = 0.043). Among participants without angina, DR was associated with a higher risk of SI (OR = 1.88), whereas among those with angina, the OR was 0.24 with a wide CI, suggesting an unstable estimate. To reduce the risk of false-positive findings due to multiple comparisons, the P values for subgroup interactions were further adjusted for multiple testing. After correction, none of the interaction P values remained statistically significant (adjusted P > .05; Table S2, Supplemental Digital Content 2), supporting the overall robustness of the association between DR and SI.
Table 3.
Subgroup analysis.
| Variable | Subgroups | Unweighted | Weighted | ||
|---|---|---|---|---|---|
| n (total) | n (event_%) | SI, OR (95% CI) | P for interaction | ||
| Age | .137 | ||||
| < 65 | Without DR | 1791 | 109 (6.1) | Reference | |
| With DR | 461 | 35 (7.6) | 1.30 (0.79–2.16) | ||
| ≥ 65 | Without DR | 1386 | 50 (3.6) | Reference | |
| With DR | 368 | 27 (7.3) | 2.64 (1.13–6.14) | ||
| Gender | .695 | ||||
| Male | Without DR | 1599 | 63 (3.9) | Reference | |
| With DR | 448 | 31 (6.9) | 1.77 (0.93–3.35) | ||
| Female | Without DR | 1578 | 96 (6.1) | Reference | |
| With DR | 381 | 31 (8.1) | 1.55 (0.85–2.83) | ||
| Hypertension | .141 | ||||
| No | Without DR | 843 | 40 (4.7) | Reference | |
| With DR | 168 | 13 (7.7) | 3.62 (1.53–8.55) | ||
| Yes | Without DR | 2334 | 119 (5.1) | Reference | |
| With DR | 661 | 49 (7.4) | 1.46 (0.89–2.39) | ||
| Congestive heart failure | .509 | ||||
| No | Without DR | 2917 | 134 (4.6) | Reference | |
| With DR | 688 | 48 (7.0) | 1.53 (0.84–2.78) | ||
| Yes | Without DR | 260 | 25 (9.6) | Reference | |
| With DR | 141 | 14 (9.9) | 3.71 (1.17–11.79) | ||
| Angina pectoris | .043 | ||||
| No | Without DR | 2968 | 141 (4.8) | Reference | |
| With DR | 747 | 58 (7.8) | 1.88 (1.16–3.07) | ||
| Yes | Without DR | 209 | 18 (8.6) | Reference | |
| With DR | 82 | 4 (4.9) | 0.24 (0.03–1.74) | ||
| Hyperlipidemia | .948 | ||||
| No | Without DR | 461 | 28 (6.1) | Reference | |
| With DR | 93 | 4 (4.3) | 1.99 (0.46–8.59) | ||
| Yes | Without DR | 2716 | 131 (4.8) | Reference | |
| With DR | 736 | 58 (7.9) | 1.70 (1.11–2.60) | ||
| Depression | .594 | ||||
| No | Without DR | 2796 | 56 (2.0) | Reference | |
| With DR | 694 | 22 (3.2) | 1.45 (0.61–3.44) | ||
| Yes | Without DR | 381 | 103 (27.0) | Reference | |
| With DR | 135 | 40 (29.6) | 2.11 (1.24–3.61) | ||
Adjusted for age, gender, race, ethnicity, marital status, education level, BMI, MET, smoking, alcohol, hypertension, congestive heart failure, angina pectoris, hyperlipidemia, and depression.
BMI = body mass index, CI = confidence interval, DR = diabetic retinopathy, MET = metabolic equivalent task, OR = odds ratio, SI = suicidal ideation.
3.3. Sensitivity analysis
To evaluate the potential impact of missing covariate data on the stability of the results, we repeated the analyses after multiple imputation (Table S3, Supplemental Digital Content 3). In the unadjusted Model I, DR was associated with higher odds of SI (OR = 1.82, 95% CI: 1.23–2.70, P = .003). With sequential adjustment for demographic characteristics (Model II), lifestyle factors (Model III), cardiovascular comorbidities (Model IV), and depressive status (Model V), the ORs were slightly attenuated but remained statistically significant. In the fully adjusted Model V, the OR was 1.59 (95% CI: 1.03–2.47, P = .037). Compared with the main analysis shown in Table 2, the effect estimates after multiple imputation were modestly lower, but the direction and statistical significance of the association were unchanged. These findings suggest that the observed association between DR and SI was not materially affected by missing covariate data.
As shown in Table S4, Supplemental Digital Content 4, further adjustment for diabetes-related covariates, including diabetes duration, HbA1c, and insulin use, slightly attenuated the association between DR and SI across the sequential models. In the fully adjusted Model V, DR remained associated with higher odds of SI (OR = 1.59, 95% CI: 1.02–2.48, P = .041). Although this estimate was modestly lower than that in the corresponding Model V in the main analysis (OR = 1.70, 95% CI: 1.08–2.68, P = .023), the direction of the association was unchanged and the CI did not cross 1. These results indicate that the association between DR and SI persisted after additional adjustment for diabetes duration, glycemic control, and insulin treatment.
We also repeated the multivariable regression analyses using a fundus examination-based definition of DR. As presented in Table S5, Supplemental Digital Content 5, after full adjustment for potential confounders, no statistically significant association was observed between DR and SI. In the severity-stratified analysis, the OR was 1.02 for mild non-proliferative DR (95% CI: 0.51–2.04, P = .958) and 0.74 for moderate-to-severe DR (95% CI: 0.26–2.12, P = .572). The proliferative DR group had no SI events, and the OR was therefore not estimable. When DR was analyzed as a binary variable, the association was also not statistically significant (OR = 1.27, 95% CI: 0.23–6.92, P = .607).
4. Discussion
In this nationally representative cross-sectional study of U.S. adults with diabetes, after adjusting for demographic characteristics, lifestyle factors, clinical comorbidities, and depressive symptom-related covariates, DR was associated with significantly higher odds of SI. Subgroup analyses showed generally consistent positive associations across most subgroups, and no subgroup interaction remained statistically significant after correction for multiple comparisons, supporting the overall robustness of the association between DR and SI. Regarding sensitivity analyses: although the association did not reach statistical significance when using a fundus examination-based definition of DR (limited to the 2005–2008 cycle, with a small sample size and a limited number of SI events), the direction of the effect remained consistent with the main analysis (OR > 1) and did not reverse. Furthermore, the positive association between DR and SI remained robust after multiple imputation for missing data and after additional adjustment for diabetes duration, HbA1c, and insulin use.
Currently, the direct evidence establishing a link between DR and SI remains limited. A hospital-based cross-sectional study among visually impaired older adults (aged ≥ 60 years) attending an eye center in Thailand reported that DR was significantly associated with SI after adjusting for multiple comorbidities (OR = 2.4, 95% CI: 1.05–5.85); the study also found a high prevalence of SI (32.5%) and suggested an interactive effect between DR and depression.[11] Our findings are consistent with this prior evidence: in a nationally representative sample of U.S. adults with diabetes, DR remained independently associated with higher odds of SI even after further adjustment for depressive symptoms (OR = 1.70, 95% CI: 1.08–2.68). Compared with the Thai study, our analysis included a broader age range of adults, applied a stricter adjustment for depressive symptoms (PHQ-8 score ≥ 10), and assessed the robustness of the association through multiple sensitivity analyses, including multiple imputation, additional adjustment for diabetes-related covariates, and use of a fundus examination-based definition of DR. Overall, our study extends the association between DR and SI from a Thai geriatric clinic-based population to a nationally representative U.S. adult diabetic population, and highlights that DR may independently increase the risk of SI even after accounting for depression.
In the multivariable regression analysis using the fundus examination-based definition of DR, the association did not reach statistical significance when DR was treated as a binary variable (OR = 1.27, 95% CI: 0.23–6.92, P = .607). This result should be interpreted cautiously because the fundus examination-based definition was available only for the NHANES 2005 to 2008 cycles. As a result, the analytic sample was substantially smaller than that in the main analysis, and only 19 SI events were observed among participants with DR. The limited number of outcome events led to a wide CI and reduced statistical precision. Although the point estimate was > 1, the CI was wide and crossed 1; therefore, this sensitivity analysis should not be interpreted as confirming the main finding. Instead, it suggests that analyses based on objective DR assessment were underpowered in the available NHANES cycles and that future studies with larger samples and clinically confirmed DR measures are needed to further evaluate this association.
The exclusion of individuals with PHQ-9 deletions might have systematically excluded a subset of high-risk individuals with a greater burden of disease and mental health risks. This form of selective exclusion likely biased our effect estimates toward 0 (i.e., underestimating the true strength of the association between DR and SI). If these excluded high-risk populations were included in the analysis, it is anticipated that the positive association between DR and SI would be stronger. Therefore, the significant positive association observed in the current analysis (OR of approximately 1.70 to 1.90 for DR and SI) was actually obtained under conservative bias, which, in turn, supports the validity of the association. It should be noted that there was no significant interaction between age and DR in the weighted subgroup analysis. Therefore, the influence of the older age of the excluded population on the interaction conclusion may be limited.
The mechanisms potentially linking DR to SI may be multifactorial, involving interconnected physiological, psychological, and social pathways. Vision loss related to DR may impair mobility and limit participation in daily activities and physical exercise,[34,35] factors that are known to be associated with depression. Progressive DR may also be a source of significant psychological distress, including depressive and anxiety symptoms.[36] Furthermore, visual impairment can reduce social engagement and increase the risk of isolation and dependency.[37] Evidence from previous studies suggests that engagement in physical activities, particularly outdoor activities, and active social interactions may reduce the risk of depression and promote psychological resilience,[38,39] whereas depressive symptoms are strongly associated with SI. In addition, diabetes- and DR-related biological changes, such as inflammation, microvascular dysfunction, and neuroendocrine dysregulation, may also contribute to psychological vulnerability, although direct evidence for these pathways remains limited. Therefore, the observed association between DR and SI may reflect a complex interaction of visual, psychological, behavioral, and biological factors. Given the cross-sectional design of this study, these proposed mechanisms remain speculative and require confirmation in future longitudinal and mechanistic studies.
Our research benefits from utilizing a comprehensive, nationally representative dataset from the U.S. population, which enhances the generalizability of our findings. We incorporated the complex survey design of NHANES, including sampling weights, stratification, and clustering, to obtain more appropriate population-level estimates. We also employed rigorous statistical adjustments for demographic, behavioral, clinical, and diabetes-related covariates and performed several sensitivity analyses to evaluate the stability of the observed association.
However, several limitations should be noted. First, this study adopted a cross-sectional design. All association analyses were unable to infer a causal relationship between DR and SI, and they could not rule out the possibility of reverse causation or residual confounding either. Second, although NHANES provides nationally representative data, the generalizability of our findings may be limited to populations with similar characteristics. Third, the self-reported DR question may miss asymptomatic or undiagnosed cases and does not capture DR severity, treatment history, or visual acuity. Fourth, SI was assessed using only a single PHQ-9 item (item 9). Although this scale has been widely used in large-scale epidemiological surveys as a screening tool for SI, its single-item design is unable to comprehensively cover the complex manifestations of SI in multiple dimensions such as frequency, intensity, subjective perception, and behavioral intention. It is particularly noteworthy that the “passive desire for death” it reflects may have an essential difference from the active SI in clinical terms and should not be regarded as the same. It may also be subject to recall bias and social desirability bias, and its psychometric performance is limited. Finally, despite adjustment for multiple demographic, lifestyle, clinical, and diabetes-related covariates, residual confounding by unmeasured factors cannot be excluded.
5. Conclusion
In conclusion, a significant and independent association was found between DR and SI, as evaluated by item 9 of the PHQ-9 score, even after adjusting for confounding factors such as depressive symptoms. This finding implies that greater attention ought to be given to the mental health of patients with DR in clinical practice, particularly for the early screening of SI. Considering the cross-sectional design of this study, no chronological or causal relationships can be deduced. Therefore, these conclusions should be interpreted with caution. In the future, prospective studies are required to further validate the stability of this association and explore its causal relationship and potential mechanisms.
Acknowledgments
This investigation was conducted using publicly accessible data from the National Health and Nutrition Examination Survey (NHANES). We express our sincere appreciation to the National Center for Health Statistics (NCHS) at the Centers for Disease Control and Prevention (CDC) for administering this comprehensive survey. Additionally, we are deeply grateful to all NHANES participants for their essential contributions to public health research.
Author contributions
Conceptualization: Najuan Huang, Yaru Du, Pengfei Wang.
Data curation: Najuan Huang, Yaru Du, Yan Zheng, Pengfei Wang.
Formal analysis: Najuan Huang, Yan Zheng, Pengfei Wang.
Investigation: Najuan Huang, Pengfei Wang.
Methodology: Yaru Du, Yan Zheng, Pengfei Wang.
Project administration: Najuan Huang, Pengfei Wang.
Resources: Najuan Huang, Yan Zheng.
Supervision: Pengfei Wang.
Validation: Pengfei Wang.
Visualization: Najuan Huang, Yaru Du, Yan Zheng.
Writing – original draft: Najuan Huang.
Writing – review & editing: Najuan Huang, Yaru Du, Yan Zheng, Pengfei Wang.
Abbreviations:
- BMI
- body mass index
- CI
- confidence interval
- DR
- diabetic retinopathy
- HbA1c
- hemoglobin A1c
- MET
- metabolic equivalent task
- NCHS
- National Center for Health Statistics
- NHANES
- National Health and Nutrition Examination Survey
- OR
- odds ratio
- PHQ
- Patient Health Questionnaire
- SI
- suicidal ideation
- U.S.
- United States
Written informed consent was secured from all individuals who participated in the NHANES survey.
The NHANES study protocol received ethical clearance from the NCHS Research Ethics Review Board, with all participants providing written informed consent. Since this study involved secondary analysis of anonymized NHANES data, our institutional review board determined that additional ethical review was unnecessary.
The authors have no funding or conflicts of interest to disclose.
The datasets generated during and/or analyzed during the current study are publicly available.
Supplemental Digital Content is available in the online version of this article (http://dx.doi.org/10.1097/MD.0000000000049694).
How to cite this article: Huang N, Du Y, Zheng Y, Wang P. Association between diabetic retinopathy and suicidal ideation among U.S. adults with diabetes: A cross-sectional NHANES study, 2005–2018. Medicine 2026;105:28(e49694).
Contributor Information
Najuan Huang, Email: najuan.h@outlook.com.
Yaru Du, Email: dyrdyr518@163.com.
Yan Zheng, Email: 344151208@qq.com.
References
- [1].Turecki G, Brent DA. Suicide and suicidal behaviour. Lancet. 2016;387:1227–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].GBD 2021 Suicide Collaborators. Global, regional, and national burden of suicide, 1990-2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Public Health. 2025;10:e189–202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Klonsky ED, May AM, Saffer BY. Suicide, suicide attempts, and suicidal ideation. Annu Rev Clin Psychol. 2016;12:307–30. [DOI] [PubMed] [Google Scholar]
- [4].Ribeiro JD, Franklin JC, Fox KR, et al. Self-injurious thoughts and behaviors as risk factors for future suicide ideation, attempts, and death: a meta-analysis of longitudinal studies. Psychol Med. 2016;46:225–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Guo S, Qing G, Yang G. The relationship between chronic disease variety and quantity and suicidal ideation: a cross-sectional study of NHANES. J Psychosom Res. 2024;184:111854. [DOI] [PubMed] [Google Scholar]
- [6].Costanza A, Vasileios C, Ambrosetti J, et al. Demoralization in suicide: a systematic review. J Psychosom Res. 2022;157:110788. [DOI] [PubMed] [Google Scholar]
- [7].Lai W, Liao Y, Zhang H, et al. The trajectory of depressive symptoms and the association with quality of life and suicidal ideation in patients with major depressive disorder. BMC Psychiatry. 2025;25:310. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Kim CY, Ha A, Shim SR, Hong IH, Chang IB, Kim YK. Visual impairment and suicide risk: a systematic review and meta-analysis. JAMA Netw Open. 2024;7:e247026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Palbo A, Roed Rasmussen ML, Hansen MS, Subhi Y. Association between visual impairment and suicide: a systematic review and meta-analysis. Acta Ophthalmol. 2024;102:263–73. [DOI] [PubMed] [Google Scholar]
- [10].Fan ZH, Xu J, Ge MW, et al. Suicide death, suicidal ideation and suicide attempt in patients with diabetes: a systematic review and meta-analysis. J Adv Nurs. 2024;80:4050–73. [DOI] [PubMed] [Google Scholar]
- [11].Tantirattanakulchai P, Hounnaklang N, Pongsachareonnont PF, Khambhiphant B, Win N, Tepjan S. Prevalence and factors associated with suicidal ideation among older people with visual impairments attending an eye center during the COVID-19 pandemic: a hospital-based cross-sectional study. Clin Ophthalmol. 2023;17:917–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Huang W, Ma X, Liang H, et al. Dietary magnesium intake affects the association between serum vitamin d and type 2 diabetes: a cross-sectional study. Front Nutr. 2021;8:763076. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Wang P, Sun X, Wang Y, Zhang Y, Liang X, Zhang L. Effects of depression and suicidal ideation on health-related quality of life in diabetic individuals: findings from a population-based study. Diabetol Metab Syndr. 2025;17:202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Li B, Zhou C, Gu C, et al. Modifiable lifestyle, mental health status and diabetic retinopathy in U.S. adults aged 18-64 years with diabetes: a population-based cross-sectional study from NHANES 1999-2018. BMC Public Health. 2024;24:11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Chen X, Sun M, Gu Z, Hao X, Xie L. Association of serum selenium levels with diabetic retinopathy: NHANES 2011-2016. Front Med (Lausanne). 2025;12:1546214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Chen WH, Hsieh KL, Chen JY, Chen CT. Association between controlling nutritional status (CONUT) score and retinopathy in adults with diabetes mellitus: an NHANES analysis. J Diabetes Res. 2025;2025:7303131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med. 2001;16:606–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [18].Martin A, Rief W, Klaiberg A, Braehler E. Validity of the brief patient health questionnaire mood scale (PHQ-9) in the general population. Gen Hosp Psychiatry. 2006;28:71–7. [DOI] [PubMed] [Google Scholar]
- [19].Costantini L, Pasquarella C, Odone A, et al. Screening for depression in primary care with Patient Health Questionnaire-9 (PHQ-9): a systematic review. J Affect Disord. 2021;279:473–83. [DOI] [PubMed] [Google Scholar]
- [20].Yu C, Hu W, Lei X. Association between cardiometabolic index (CMI) and suicidal ideation: the mediating role of depression and cardiovascular disease. J Affect Disord. 2025;380:487–95. [DOI] [PubMed] [Google Scholar]
- [21].Simon GE, Rutter CM, Peterson D, et al. Does response on the PHQ-9 depression questionnaire predict subsequent suicide attempt or suicide death? Psychiatr Serv. 2013;64:1195–202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Rossom RC, Coleman KJ, Ahmedani BK, et al. Suicidal ideation reported on the PHQ9 and risk of suicidal behavior across age groups. J Affect Disord. 2017;215:77–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Moazzami K, Dolmatova EV, Feurdean M. Suicidal ideation among adults with cardiovascular disease: the national health and nutrition examination survey. Gen Hosp Psychiatry. 2018;51:5–9. [DOI] [PubMed] [Google Scholar]
- [24].Negeri ZF, Levis B, Sun Y, et al. Accuracy of the patient health questionnaire-9 for screening to detect major depression: updated systematic review and individual participant data meta-analysis. BMJ. 2021;375:n2183. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Diep C, Bhat V, Wijeysundera DN, Clarke HA, Ladha KS. The Association between recent cannabis use and suicidal ideation in adults: a population-based analysis of the NHANES from 2005 to 2018. Can J Psychiatry. 2022;67:259–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Huang H, Fu J, Lu K, Fu Y, Zhuge P, Yao Y. Association between dietary fiber intake and suicidal ideation: a cross-sectional survey. Front Nutr. 2024;11:1465736. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Cui X, Wen D, Xiao J, Li X. The causal relationship and association between biomarkers, dietary intake, and diabetic retinopathy: insights from Mendelian randomization and cross-sectional study. Diabetes Metab J. 2025;49:1087–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Liang J, Huang S, Jiang N, et al. Association between joint physical activity and dietary quality and lower risk of depression symptoms in US adults: cross-sectional NHANES study. JMIR Public Health Surveill. 2023;9:e45776. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Chen L, Cai M, Li H, et al. Risk/benefit tradeoff of habitual physical activity and air pollution on chronic pulmonary obstructive disease: findings from a large prospective cohort study. BMC Med. 2022;20:70. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30].Rattan P, Penrice DD, Ahn JC, et al. Inverse association of telomere length with liver disease and mortality in the US population. Hepatol Commun. 2022;6:399–410. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [31].Zhang Z, Wang P, Cui G, Li H. Higher HEI-2015 score is associated with reduced risk of fecal incontinence: insights from a large cross-sectional study. BMC Public Health. 2024;24:3221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [32].Blighe K, Gurudas S, Lee Y, Sivaprasad S. Diabetic retinopathy environment-wide association study (EWAS) in NHANES 2005-2008. J Clin Med. 2020;9:3643. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [33].Li H, Wang B, Liu C. Association between age at diabetes diagnosis and the development and progression of diabetic retinopathy. Sci Rep. 2025;15:13827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [34].Ong SR, Crowston JG, Loprinzi PD, Ramulu PY. Physical activity, visual impairment, and eye disease. Eye (Lond). 2018;32:1296–303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [35].Cai Y, Schrack JA, Wang H, et al. Visual impairment and objectively measured physical activity in middle-aged and older adults. J Gerontol A Biol Sci Med Sci. 2021;76:2194–203. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [36].Khoo K, Man REK, Rees G, Gupta P, Lamoureux EL, Fenwick EK. The relationship between diabetic retinopathy and psychosocial functioning: a systematic review. Qual Life Res. 2019;28:2017–39. [DOI] [PubMed] [Google Scholar]
- [37].Zheng DD, Lee DJ, Rundek T, et al. Visual impairment and cognitive function in aging adults: sex and age differences in mediating effect of social isolation and depression. Am J Ophthalmol. 2025;274:196–208. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [38].Singh B, Olds T, Curtis R, et al. Effectiveness of physical activity interventions for improving depression, anxiety and distress: an overview of systematic reviews. Br J Sports Med. 2023;57:1203–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [39].Domènech-Abella J, Lara E, Rubio-Valera M, et al. Loneliness and depression in the elderly: the role of social network. Soc Psychiatry Psychiatr Epidemiol. 2017;52:381–90. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.