Posttraumatic Stress Disorder and Type 2 Diabetes Outcomes in Veterans

Key Points Question What is the association between meeting diagnostic criteria for posttraumatic stress disorder (PTSD) and risk of poor type 2 diabetes (T2D) outcomes? Findings In this cohort study of 10 002 veterans, no longer meeting diagnostic criteria for PTSD was associated with a lower risk of microvascular complications. Among veterans aged 18 to 49 years, but not among those aged 50 to 80 years, no longer meeting PTSD criteria was associated with a lower likelihood of starting insulin and a lower risk of all-cause mortality. Meaning The findings of this study suggest that PTSD is a modifiable risk factor for some adverse T2D outcomes among patients with comorbid PTSD and T2D.


Introduction
Individuals with psychiatric disorders have a 20-year shorter life expectancy than those without mental illness, and despite advances in treatment, there has been little to no narrowing of the mental health disparity in mortality over the past 30 years. 1,2This disparity in mortality may be related to the association between mental illness and poor metabolic health.[13][14] Some evidence suggests that PTSD might be a modifiable risk factor for T2D and adverse T2D outcomes.[17] Improvement of PTSD in comorbid psychiatric illness is associated with better overall well-being 18,19 and with lower risk of some chronic health conditions, including T2D. 20 To our knowledge, there is no existing literature on T2D outcomes after PTSD improvement among patients with comorbid PTSD and T2D.This study aimed to help fill this gap in the literature.
If there is a reduced risk of adverse T2D outcomes among patients with PTSD who experience improvement and no longer meet PTSD criteria, then findings could be used to incentivize PTSD treatment and patients should be educated that improvement in PTSD may benefit comorbid T2D management.We investigated whether veterans with comorbid PTSD and T2D who no longer meet PTSD diagnostic criteria had a lower risk of starting insulin (a proxy for worsening T2D control), poor glycemic control, microvascular complications, or all-cause mortality compared with those with persistent PTSD.We planned subgroup analyses to determine whether age group, sex, race, PTSD severity, and comorbid depression status were significant effect modifiers.

Methods
This cohort study used data from the US Veterans Health Administration (VHA).Because the data were deidentified and the investigators could not reidentify patients, the Saint Louis University and VHA institutional review boards deemed this study exempt and informed consent was waived.The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Study Population
All variables were created from deidentified VHA administrative medical record data from October  no more than 12 months after the first PCL score, to assess for symptom improvement in the exposure year. 21,22llowing methods established by Moshier et al, 23 PCL-4 scores were cross-walked to PCL-5 scores.To create an exposure year to measure PCL change and comorbid T2D, PCL scores of 33 or greater were measured between FY2013 and FY2019.Type 2 diabetes was defined with ICD-9 (250.x0 and 250.x2) or ICD-10 (E11*) codes.The end of this exposure year was the index (ie, baseline) date, which could occur between FY2014 and FY2020; this allowed for a 2-year look-back to measure covariates.This was a dynamic cohort, meaning that patients entered the study whenever they met the eligibility criteria.All eligible patients had a possible 2 to 9 years of follow-up.
The index date was the end of an exposure year that met the following criteria: (1) at least 1 PCL score 8 weeks after and within 12 months of the first PCL score of 33 or greater; (2) a T2D diagnosis coded in the exposure year; and (3) no insulin fills, type 1 diabetes, or microvascular complications in the 2 years before the end of the exposure year.After defining each patient's index year, we then limited the sample to patients aged 18 to 80 years who had reasonably well-controlled hemoglobin A 1c (HbA 1c ) at index date (ie, last value in the year before index date <7.5%). 24Finally, those with missing demographic information were excluded, leaving an analytic sample of 10 002 patients diagnosed with PTSD who either met or no longer met the criteria for PTSD at the end of the exposure year.The sampling approach is illustrated in Figure 1.The study design is illustrated in

Study Variables
Detailed variable definitions are presented in eTable 1 in Supplement 1. Type 2 diabetes outcomes were insulin initiation, poor glycemic control, any microvascular complication, and all-cause mortality.Insulin has traditionally been used as augmentation therapy, and we treated insulin initiation as a marker of poor T2D control.Recommendations on glycemic control vary and HbA 1c goals are based on patient age, comorbidities, presence of microvascular complications, and treatment preferences.Tight control with a goal HbA 1c of less than 7% is recommended by the American Diabetes Association and the VHA Department of Defense clinical practice guideline for many nonpregnant adults. 24,25However, for patients with limited life expectancy, increased risk of hypoglycemic complications, or both, HbA 1c goals are more liberal.Therefore, we used a cutoff HbA 1c of 7.5% or less to define good control, and we considered an HbA 1c greater than 7.5% during follow-up as poor glycemic control. 24,25y microvascular complication was defined by the presence of ICD-

Exposure
At the start of the exposure year, all patients had a PCL score of 33 or greater.Scores ranging from 31 to 33 are appropriate thresholds for identifying current PTSD. 22Patients were classified as no longer meeting PTSD criteria at index date if their last PCL score in the exposure year (using the aforementioned approach) was less than 33.Patients still met PTSD criteria if their PCL score remained at 33 or greater.

Covariates
Unless otherwise indicated, covariates other than demographic factors were measured in the 2 years before index date.We controlled for index fiscal year and the following demographic characteristics: the distribution of average outpatient clinic visits per month per patient.We controlled for access to non-US Department of Veterans Affairs health insurance as a proxy for socioeconomic status.
Because the likelihood of a score falling below the PCL threshold of 33 partly depends on severity at index date, we controlled for very severe PTSD, which we defined as a PCL score of 66 or greater at the start of the exposure year.The following psychiatric comorbidities were controlled for: depression, dysthymia, any anxiety disorder (ie, social phobia, panic disorder, generalized anxiety disorder, or anxiety disorder not otherwise specified), obsessive-compulsive disorder, bipolar disorder, schizophrenia, alcohol or drug abuse or dependence, and smoking or nicotine dependence.
To investigate associations with PTSD improvement that could occur with or without treatment, we controlled for minimally adequate PTSD psychotherapy (at least 9 sessions in any 15-week period).We did so because patients who remain in treatment may be more likely to engage in other healthy behaviors that could reduce the risk of T2D complications compared with those who do not.
We controlled for other psychotropic therapies, including 12 or more weeks of antidepressant therapy (ie, at least acute-phase treatment).[28] Physical comorbidities included metabolic and cardiovascular conditions posited to be more common in PTSD and to contribute to worse T2D outcomes.These conditions included obesity, hyperlipidemia, atrial fibrillation, angina, heart failure, left ventricular hypertrophy, myocardial infarction, peripheral vascular disease, and stroke.We controlled for HbA 1c at index date, which was the last HbA 1c measurement available in the year before index date.Finally, we controlled for sustained use of antidiabetic medications (metformin, sulfonylurea, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide 1 agonists, sodium-glucose cotransporter 2 inhibitors, and thiazolidinediones).

Control for Confounding
We controlled for confounding with entropy balancing to remove differences in the distribution of covariates between patients whose PTSD persisted vs those who no longer met PTSD criteria. 29,30In this way, balancing emulates randomization.Entropy balancing weights were computed using the WeightIt package in R, version 4.2.1 (R Project for Statistical Computing). 31The percentage of standardized mean difference (SMD% = 100 × SMD) was computed to evaluate balance.Wellbalanced covariates have an SMD percentage of less than 10%. 32This balancing approach to mitigate confounding permits analysis of the total average treatment effect of no longer meeting PTSD criteria on risk of adverse T2D outcomes.
Bivariate comparisons between covariates and whether patients continued to meet PTSD criteria were evaluated using χ 2 tests or independent samples t tests.We used SMD percentages to measure balance before and after entropy balancing.Competing risk survival models, 33 which account for the competing risk of death, were used to estimate the associations between no longer   32 The proportional hazards assumption was tested and met for all models (P > .10).

Subgroup Comparisons
In previous studies, younger patients (18-50 vs >50 years) tended to have worse glycemic control, 34 Black individuals tended to have more diabetes complications than White patients, 34 and T2D distress was more common among female individuals compared with male indviduals. 35Comorbid depression has also been associated with poor T2D outcomes. 36Finally, more severe PTSD is linked to increased risk of microvascular complications. 37Therefore, planned subgroup analyses were performed for age group (18-49 vs 50-80 years), race (Black, White, or other), sex (male or female), comorbid depression status, and PTSD severity at the start of the exposure year (PCL score indicating severe PTSD: Ն66 vs <66).Entropy balancing was computed for each subgroup analysis.An interaction term for whether patients continued to meet PTSD criteria and each stratification variable assessed whether effect modification was present (P < .05indicated significant modification).

Post Hoc Sensitivity Analysis
It is possible that a PCL score of less than 33 is not sufficient improvement to result in a change in physiology or in health behaviors associated with better T2D outcomes.To explore this issue, we conducted post hoc analyses using additional PCL score thresholds for change in PTSD severity.We compared risk of T2D outcomes among patients with PCL scores of 18 or less (reference group), 33 to 65, or 66 to 80. Separate sensitivity analyses modeled poor glycemic control as an HbA 1c greater than 7.0%.Finally, we adjusted for use of the VHA's evidence-based weight loss program MOVE! 38 because PTSD improvement may be linked to adopting healthier behavior and fewer adverse T2D outcomes.

Results
This study included 10 002 veterans.Of these patients, 65.3% were aged older than 50 years, 61.4% were married, 87.2% were men, and 12.8% were women (Table 1  Both before and after controlling for confounding, the incidence of starting insulin, poor glycemic control, or mortality did not significantly differ between patients who no longer met criteria for PTSD compared with those with persistent PTSD (Table 3).After controlling for confounding using weighted data, no longer meeting criteria for PTSD was associated with a significantly lower risk of microvascular complications (hazard ratio, 0.92 [95% CI, 0.85-0.99]).
Subgroup analyses revealed statistically significant interactions by age group and meeting PTSD criteria and risk of insulin initiation and all-cause mortality (eTable 2 in Supplement 1).Among patients aged 18 to 49 years, but not among those aged 50 to 80 years, no longer meeting PTSD criteria was associated with a lower likelihood of starting insulin (HR, 0.69 [95% CI, 0.53-0.88];P = .003)and a lower risk of all-cause mortality (HR, 0.39 [95% CI, 0.19-0.83];P = .008).
The association between no longer meeting PTSD criteria and T2D outcomes did not significantly differ by sex (eTable 3 in Supplement 1), by race (eTable 4 in Supplement 1), or by severe PTSD (eTable 6 in Supplement 1) at index date.However, there was a significant interaction between depression and no longer meeting PTSD criteria and risk of starting insulin.Among those without depression, no longer meeting PTSD criteria was associated with a lower risk of insulin initiation (HR, 0.73 [95% CI, 0.55-0.97];P = .03)(eTable 5 in Supplement 1).
Sensitivity analyses (eTable 7 in Supplement 1) using different PCL thresholds at index date to define exposure groups revealed that compared with a PCL score of 18 or less at index date, the risk of microvascular complications was significantly greater for those with a PCL score of 19 to 32 (HR, 1.23 [95% CI, 1.06-1.42])and nearly identical point estimates were observed for PCL scores of 33 to 65 and 66 to 80. Thus, there appeared to be a greater magnitude of reduced risk of microvascular complications among those with PCL scores of 18 or less.There were no associations between these other PCL thresholds and insulin initiation, glycemic control, and all-cause mortality.
Results did not change in sensitivity analyses treating poor glycemic control as an HbA 1c greater than 7.0%.Finally, results did not change after adjusting for use of the MOVE! program.

Discussion
In this cohort study of VHA patients with comorbid PTSD and T2D, we observed that no longer meeting diagnostic criteria for PTSD was associated with a modest (8%) reduction in risk of microvascular complications, compared with continuing to meet PTSD criteria.Among patients aged 18 to 49 years (but not among older patients), no longer meeting PTSD criteria was associated with a lower risk of insulin initiation and all-cause mortality.It is logical that younger, adult VHA patients would have a reduced risk of insulin initiation and mortality because the accumulation of comorbid Michopoulos et al 39 described PTSD as a metabolic disease with common underlying mechanisms related to the inflammatory response.From this perspective, it is possible that physiological abnormalities in the hypothalamic-pituitary-adrenal axis, changes in metabolic hormones, and poor diet and lack of exercise explain why PTSD is a risk factor for prediabetes and T2D.Although associated with incident T2D, the present findings, along with sparse existing literature, 40 suggest that large decreases in PTSD severity may have a modest favorable association with some T2D outcomes, particularly microvascular complications.In younger age groups and among those without comorbid depression, the magnitude of associations was much greater.
Clinicians managing T2D among patients with PTSD should ensure that evidence-based treatments are used to reduce PTSD symptoms and improve T2D outcomes, particularly among younger individuals.
We observed that patients without depression and no longer meeting PTSD criteria were statistically significantly less likely to start insulin.2][13] In addition, this finding is aligned with the link between depression and hyperglycemia 41 and is consistent with evidence that depression and correlated inflammation and oxidative stress increase risk of poor T2D outcomes. 36,42This finding could also be related to poor diabetes self-management and poor medication adherence among persons with PTSD and depression. 42 a 2023 study of PTSD severity in patients with comorbid T2D, we observed that severe PTSD, compared with mild PTSD, was associated with an increased risk of microvascular complications but not with poor glycemic control and insulin initiation. 37The lack of associations between PTSD status and glycemic control agrees with a prior study demonstrating that HbA 1c values in comorbid PTSD and T2D do not meaningfully differ between groups with both PTSD and depression, with PTSD alone, with depression alone, and without PTSD and depression. 43Another explanation for good glycemic control is more frequent glucose self-monitoring, which has been observed in individuals with more severe PTSD. 44The lack of association between PTSD status and glycemic control might be related to the VHA's excellent T2D management relative to the private sector. 45The VHA outperforms the civilian sector in T2D treatment, with a much higher percentage of VHA patients achieving control than the private sector. 45It is possible the increases in insulin use that we observed in some subgroups reflect intensification of medications that prevented patients from crossing thresholds for poor control.With such well-managed T2D, we may have lacked sufficient variability in HbA 1c values in follow-up to detect an association with PTSD improvement.

Limitations
This study had some limitations.We do not know how long the patients in this cohort had comorbid PTSD and T2D.Heterogeneity in duration of these conditions could bias our results.We could not overcome this limitation because there were too few eligible patients to form a cohort of patients with new-onset PTSD and newly diagnosed T2D.Yet such a cohort would identify an atypical pattern of onset that would not represent the majority of patients with comorbid PTSD and T2D.This study is based on medical record data from the VHA and results may not generalize to non-VHA health systems.Additional research in more diverse samples is needed.Misclassification is a risk in retrospective cohort designs, particularly when relying on historical medical record data.For example, we might have biased analyses toward the null if we substantially misclassified patients as unaffected if they had undiagnosed microvascular complications.We controlled for a large number of potential confounders but we are not able to rule out unmeasured confounding.The VHA's superior diabetes management may have reduced our ability to detect adverse T2D outcomes.The majority of the sample was male and results may not generalize to female individuals with PTSD and T2D.
1, 2011, to September 30, 2022 (fiscal years FY2012-FY2022).Medical record data included diagnostic codes (International Classification of Diseases, Ninth Revision [ICD-9], and Tenth Revision [ICD-10]), Current Procedural Terminology codes, pharmacy records, laboratory results, vital signs, vital status, repeated PTSD Checklist (PCL) scores, and demographic information.Eligibility A base sample of patients included those with at least 1 PTSD diagnosis (ICD-9 code 309.81 or ICD-10 code F43.1*) between FY2012 and FY2022.Potentially eligible patients had at least 1 PCL score (according to PTSD diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders, JAMA Network Open | Psychiatry

Network Open | Psychiatry
28 856 With T2D in exposure year (year after PCL score ≥33) 21 126 Without T1D or insulin fills 2 y before index date (ie, end of exposure year) 12 909 With HbA 1c available for year before index date and at follow-up 10 276 With index HbA 1c (last in year before index date) <7.5% 10 255 Aged 18-80 y at index date 10 002 With complete demographic data eligible for analysis 14 815 With no microvascular complications 2 y before index date (1 y + first PCL score meeting prior criteria) 6311 Excluded (multiple reasons possible) 2279 With diabetic nephropathy 4080 With diabetic neuropathy 1396 With diabetic retinopathy 253 Excluded (multiple reasons possible) 185 Missing region 13 Missing race 58 Missing marital status 7730 Excluded (multiple reasons possible) 7219 With insulin fills 1788 With T1D 1906 Excluded (multiple reasons possible) 1586 No HbA 1c 1 y before index date 527 No HbA 1c at follow-up 163 436 Excluded (no T2D in exposure year) 2633 Excluded (HbA 1c ≥7.5%) 21 Excluded (not aged 18-80 y) FY indicates fiscal year; HbA 1c , hemoglobin A 1c ; PCL, PTSD Checklist; PTSD, posttraumatic stress disorder; T1D, type 1 diabetes; T2D, type 2 diabetes.
Posttraumatic Stress Disorder and Type 2 Diabetes Outcomes in Veterans meeting PTSD criteria and each T2D outcome.A Cox proportional hazards regression model was computed to estimate risk of all-cause mortality.The censor date for starting insulin and for microvascular complications was either death or last visit date in follow-up; for poor glycemic control, it was death or last HbA 1c measurement.The censor date for the all-cause mortality model was the end of follow-up (September 30, 2022).Follow-up time was calculated as months from index date to either the outcome date or the censor date.All models were fitted before and after entropy balance weighting to calculate hazard ratios (HRs) and 95% CIs.Weighted models used robust, sandwichtype variance estimators for CIs.

Table 1 .
Baseline Characteristics of Patients With Comorbid PTSD and Type 2 Diabetes by PTSD Status a for PTSD, the median time between first and last scores was 26 (IQR,15-40)weeks.Table 1 presents the covariate distributions, before balancing, among patients who did vs did not continue to meet PTSD criteria.Patients who no longer met PTSD criteria were more likely to be aged older than 60 years (SMD, 19.6%).Severe PTSD (SMD, 39.4%) and atypical antipsychotic use (SMD, 13.8%) were more prevalent among patients who continued to meet PTSD criteria.After JAMA Network Open | Psychiatry JAMA Network Open.2024;7(8):e2427569.doi:10.1001/jamanetworkopen.2024.27569(Reprinted) August 13, 2024 7/14 Downloaded from jamanetwork.comby guest on 09/17/2024 weighting the data, all covariates balanced between those who did and did not meet PTSD criteria (weighted SMD < 10%).person-years), poor glycemic control (137.1 vs 133.7 per 1000 person-years), any microvascular complication (108.4 vs 104.8 per 1000 person-years), and all-cause mortality (11.2 vs 11.0 per 1000 person-years) compared with those with persistent PTSD.

Table 1 .
Baseline Characteristics of Patients With Comorbid PTSD and Type 2 Diabetes by PTSD Status a (continued) a Unless indicated otherwise, values are presented as the No. (%) of patients.bDefined as American Indian or Alaska Native, Asian, or Native Hawaiian or Other Pacific Islander.cFirst PCL score of 66 or greater.

Table 2 .
Type 2 Diabetes Outcomes: Cumulative Incidence Percent and Incidence Rate by PTSD Status

Table 3 .
Association Between PTSD Status and Type 2 Diabetes Outcomes Before and After Entropy Balance Weighting to Control for Confounding conditions that contribute to poor health outcomes increases with aging and likely limits the ability to detect associations between PTSD improvement and T2D outcomes in those aged older than 50 years.