PTSD and Incident Heart Failure Among US Veterans
PTSD and Incident Heart Failure Among US Veterans
The Veterans Affairs Pacific Islands Health Care System (VAPIHCS) spans an area of 4.8 million square miles, from Hawaii in the east to Guam in the west, and serves approximately 130 000 veterans. We first examined VAPIHCS outpatient administrative records to assemble an appropriate cohort of eligible veterans for study. We identified all veterans who had received care prior to January 1, 2002, yielding 43 286 records. We then identified veterans with more than 1 visit between January 1, 2002, and December 31, 2004. We queried data from this period for 2 reasons: (1) to ensure that study participation was restricted to veterans receiving continuous care in the VAPIHCS system prior to the study start and (2) to ascertain baseline assessment data. We used the 2-year assessment period to determine each eligible veteran's most proximal VAPIHCS visit before our study start date of January 1, 2005; data from this visit were our baseline data. Among this group, we included individuals who were at least 45 years of age as of January 1, 2005, but no older than 89 years at the study's conclusion (December 31, 2012). To ensure that all cases of heart failure in our study were incident cases, we excluded all veterans with a diagnosis of heart failure in the VAPIHCS record as of baseline assessment. To prevent inclusion of veterans who already had heart failure before they developed PTSD, we excluded any veteran with an heart failure diagnosis antedating the PTSD diagnosis. We employed an 8-year follow-up period (January 1, 2005 to December 31, 2012) for the ascertainment of heart failure.
Application of these inclusion and exclusion criteria resulted in a preliminary sample of 11 864 veterans (Figure 1). We further excluded 3616 veterans with missing values for body mass index (BMI; defined as weight in kilograms divided by height in meters squared), because BMI has been consistently shown to predict heart failure. Applying these exclusion criterion yielded a final sample of 8248 veterans (Figure 1).
(Enlarge Image)
Figure 1.
Participant flow diagram of US veterans seen at the Veterans Affairs Pacific Islands Health Care System, 2005–2012.
Note. BMI = body mass index; HF = heart failure; VAPIHCS = Veterans Affairs Pacific Islands Health Care System.
The primary outcome was incident heart failure, operationalized as the first heart failure diagnosis recorded in the veteran's administrative record. We used International Classification of Diseases, Ninth Revision (ICD-9) codes to define an outpatient diagnosis of heart failure (Table A, available as a supplement to the online version of this article at http://www.ajph.org). To classify PTSD status, we obtained VAPIHCS records of all outpatient visits yielding a PTSD diagnostic code (ICD-9 309.81), and we considered the first occurrence as a proxy for PTSD diagnosis date. We identified comorbid conditions from the outpatient administrative record at baseline (most proximal VAPIHCS visit before January 1, 2005). We included conditions that were (1) associated with heart failure (age, BMI, diabetes, hyperlipidemia, and hypertension), (2) associated with PTSD (depression adjustment disorder and anxiety disorder), and (3) associated with both heart failure and PTSD (tobacco use and substance use disorder). (Associated diagnosis codes are listed in Table A.) Using American Heart Association guidelines, we classified veterans as normal weight (BMI < 25.0), overweight (BMI = 25.0–29.9), or obese (BMI ≥ 30.0). We additionally collected information on veterans' medications for comorbid conditions as of study baseline (for related medication classes, see Table A). We categorized medications by the VA Drug Class Code.
We classified veterans as married or not married on the basis of the most recent marital status indicated in the medical record. We used marital status as a proxy for social support, given increasing evidence for the role of sociodemographic variables in both PTSD and CHD. We also assessed variables related to veterans' military time, such as combat service (yes or no) and period of military service. Combat service has been shown to be related to PTSD in veterans, and we used period of military service to further characterize the population. We classified veterans chronologically by service period; the earliest conflict represented in our sample was World War II and the most recent was the Persian Gulf Wars.
We calculated time at risk for heart failure as of January 1, 2005. We estimated hazard ratios and 95% confidence intervals for the association between PTSD and heart failure using a Cox proportional hazards model. We censored observations at the time of an heart failure diagnosis, at the time of death, or at study end for those who did not develop heart failure or die. We assessed the proportional hazards assumption for each predictor to determine that the relative risk of heart failure between veterans with PTSD and veterans without PTSD remained constant over time. We used backward selection for multivariable regression, with a P value cutoff of .05 for variable selection. We chose backward selection over more manual variable inclusion methods in an attempt to isolate significant associations with heart failure and evaluate the effect of PTSD within that framework. We evaluated tobacco use and substance use as potential effect measure modifiers in the analyses, as they can be both a predictor and outcome (for associated ICD-9 codes, see Table A). We created interaction terms and used a likelihood ratio test to assess the statistical significance of the interaction terms, and used a P value cutoff of .2 for variable selection.
We conducted 3 sensitivity analyses to assess the limitations of our data. The first was to determine the effect of a PTSD diagnosis during the study period. Although the number of veterans with PTSD diagnosed after baseline was a small fraction of the total sample (5%), these veterans represented roughly one quarter of the PTSD group. Thus, we excluded veterans with PTSD diagnoses occurring after the study baseline to determine whether there was a substantive change in the direction or magnitude of the observed associations. Next, we ran a posthoc analysis with a broader definition for comorbid illnesses. To achieve this, we identified the comorbid illnesses (i.e., diabetes, hypertension, hyperlipidemia, depression, anxiety disorder, and adjustment disorder) by the presence of either a diagnosis in the outpatient medical record or pharmaceutical treatment of each illness. This allowed us to examine our data in a more clinically relevant, yet less specific, context. We assessed this sensitivity analysis for a meaningful change in the direction or magnitude of the observed associations. Finally, we conducted analyses to determine the potential effect of multicollinearity between predictive variables.
Methods
The Veterans Affairs Pacific Islands Health Care System (VAPIHCS) spans an area of 4.8 million square miles, from Hawaii in the east to Guam in the west, and serves approximately 130 000 veterans. We first examined VAPIHCS outpatient administrative records to assemble an appropriate cohort of eligible veterans for study. We identified all veterans who had received care prior to January 1, 2002, yielding 43 286 records. We then identified veterans with more than 1 visit between January 1, 2002, and December 31, 2004. We queried data from this period for 2 reasons: (1) to ensure that study participation was restricted to veterans receiving continuous care in the VAPIHCS system prior to the study start and (2) to ascertain baseline assessment data. We used the 2-year assessment period to determine each eligible veteran's most proximal VAPIHCS visit before our study start date of January 1, 2005; data from this visit were our baseline data. Among this group, we included individuals who were at least 45 years of age as of January 1, 2005, but no older than 89 years at the study's conclusion (December 31, 2012). To ensure that all cases of heart failure in our study were incident cases, we excluded all veterans with a diagnosis of heart failure in the VAPIHCS record as of baseline assessment. To prevent inclusion of veterans who already had heart failure before they developed PTSD, we excluded any veteran with an heart failure diagnosis antedating the PTSD diagnosis. We employed an 8-year follow-up period (January 1, 2005 to December 31, 2012) for the ascertainment of heart failure.
Application of these inclusion and exclusion criteria resulted in a preliminary sample of 11 864 veterans (Figure 1). We further excluded 3616 veterans with missing values for body mass index (BMI; defined as weight in kilograms divided by height in meters squared), because BMI has been consistently shown to predict heart failure. Applying these exclusion criterion yielded a final sample of 8248 veterans (Figure 1).
(Enlarge Image)
Figure 1.
Participant flow diagram of US veterans seen at the Veterans Affairs Pacific Islands Health Care System, 2005–2012.
Note. BMI = body mass index; HF = heart failure; VAPIHCS = Veterans Affairs Pacific Islands Health Care System.
Measures
The primary outcome was incident heart failure, operationalized as the first heart failure diagnosis recorded in the veteran's administrative record. We used International Classification of Diseases, Ninth Revision (ICD-9) codes to define an outpatient diagnosis of heart failure (Table A, available as a supplement to the online version of this article at http://www.ajph.org). To classify PTSD status, we obtained VAPIHCS records of all outpatient visits yielding a PTSD diagnostic code (ICD-9 309.81), and we considered the first occurrence as a proxy for PTSD diagnosis date. We identified comorbid conditions from the outpatient administrative record at baseline (most proximal VAPIHCS visit before January 1, 2005). We included conditions that were (1) associated with heart failure (age, BMI, diabetes, hyperlipidemia, and hypertension), (2) associated with PTSD (depression adjustment disorder and anxiety disorder), and (3) associated with both heart failure and PTSD (tobacco use and substance use disorder). (Associated diagnosis codes are listed in Table A.) Using American Heart Association guidelines, we classified veterans as normal weight (BMI < 25.0), overweight (BMI = 25.0–29.9), or obese (BMI ≥ 30.0). We additionally collected information on veterans' medications for comorbid conditions as of study baseline (for related medication classes, see Table A). We categorized medications by the VA Drug Class Code.
We classified veterans as married or not married on the basis of the most recent marital status indicated in the medical record. We used marital status as a proxy for social support, given increasing evidence for the role of sociodemographic variables in both PTSD and CHD. We also assessed variables related to veterans' military time, such as combat service (yes or no) and period of military service. Combat service has been shown to be related to PTSD in veterans, and we used period of military service to further characterize the population. We classified veterans chronologically by service period; the earliest conflict represented in our sample was World War II and the most recent was the Persian Gulf Wars.
Time-to-event Analyses
We calculated time at risk for heart failure as of January 1, 2005. We estimated hazard ratios and 95% confidence intervals for the association between PTSD and heart failure using a Cox proportional hazards model. We censored observations at the time of an heart failure diagnosis, at the time of death, or at study end for those who did not develop heart failure or die. We assessed the proportional hazards assumption for each predictor to determine that the relative risk of heart failure between veterans with PTSD and veterans without PTSD remained constant over time. We used backward selection for multivariable regression, with a P value cutoff of .05 for variable selection. We chose backward selection over more manual variable inclusion methods in an attempt to isolate significant associations with heart failure and evaluate the effect of PTSD within that framework. We evaluated tobacco use and substance use as potential effect measure modifiers in the analyses, as they can be both a predictor and outcome (for associated ICD-9 codes, see Table A). We created interaction terms and used a likelihood ratio test to assess the statistical significance of the interaction terms, and used a P value cutoff of .2 for variable selection.
We conducted 3 sensitivity analyses to assess the limitations of our data. The first was to determine the effect of a PTSD diagnosis during the study period. Although the number of veterans with PTSD diagnosed after baseline was a small fraction of the total sample (5%), these veterans represented roughly one quarter of the PTSD group. Thus, we excluded veterans with PTSD diagnoses occurring after the study baseline to determine whether there was a substantive change in the direction or magnitude of the observed associations. Next, we ran a posthoc analysis with a broader definition for comorbid illnesses. To achieve this, we identified the comorbid illnesses (i.e., diabetes, hypertension, hyperlipidemia, depression, anxiety disorder, and adjustment disorder) by the presence of either a diagnosis in the outpatient medical record or pharmaceutical treatment of each illness. This allowed us to examine our data in a more clinically relevant, yet less specific, context. We assessed this sensitivity analysis for a meaningful change in the direction or magnitude of the observed associations. Finally, we conducted analyses to determine the potential effect of multicollinearity between predictive variables.