Paul A. Heidenreich, MD, MS

 Hospital Compare, a website maintained by CMS, allows comparisons of outcomes and processes of care but not of patient satisfaction for hospitals within the Veteran Affairs (VA) Healthcare System. Therefore, we sought to compare online hospital ratings between VA hospitals and their local affiliated hospitals. 

 We identified 39 VA hospitals and a non-VA affiliated hospital with at least 2 online Yelp ratings. We determined the difference in the mean rating (VA-affiliate rating) with weighting by the number of ratings for each hospital. We used multivariate regression to compare mean Yelp ratings between VA and non-VA affiliate hospitals, adjusting for hospital characteristics (bed size, teaching status, and accreditation). 

The mean patient rating for VA hospitals (± standard deviation) was higher (3.64 ± 1.0) than the rating for affiliated hospitals (3.09 ± 0.8; 

= .0036). There was no significant correlation in rating between a VA hospital and its affiliate (r = 0.07; 

= .59). After adjustment for hospital characteristics, the adjusted rating difference (VA-affiliate hospitals) was 0.65 (95% confidence interval, 0.18-1.12). 

VA hospitals had higher patient ratings than their non-VA affiliated hospitals, a finding not explained by bed size or teaching status.

Patient satisfaction is an important measure of clinical care but cannot be compared between Veterans Affairs (VA) and non-VA hospitals using government data (eg, Hospital Compare).

Online patient ratings of hospitals (eg, Yelp) have been shown to correlate strongly with more conventional measures of patient satisfaction, as well as patient outcome.

VA hospitals had significantly higher Yelp ratings than non—VA-affiliated hospitals suggesting better patient satisfaction at VA hospitals.

The results were not explained by differences in hospital characteristics, such as size and teaching status.

Recently, the Department of Veterans Affairs (VA) started the Choice Program (section 101 of the Veterans Access, Choice, and Accountability Act of 2014), which pays for veterans to receive care at non-VA hospitals if care cannot be provided in a timely manner (within 30 days) at a VA hospital or if the patient lives a significant distance (greater than 40 miles) from one.

 Hospital Compare, a website maintained by CMS, allows comparisons of outcome and processes of care for VA and non-VA hospitals.

 Patient satisfaction is reported on Hospital Compare for non-VA hospitals only. Currently, the only way for patients to compare VA and non-VA hospitals on patient experience is through private online customer rating services (eg, Yelp). 

A recent study found a high correlation between Yelp ratings and the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) scores as reported by Medicare on Hospital Compare for non-VA US hospitals.

 Another comparison of HCAHPS and Yelp ratings for non-VA US hospitals found that Yelp ratings covered the main domains of HCAHPS while providing information on additional domains of patient experience.

 These data suggest that Yelp may be a reasonable alternative for estimates of the patient experience at VA hospitals. 

A common alternative to a VA hospital is an affiliated university hospital. Over 90% of VA hospitals are affiliated with medical schools and many share trainees and physician staff with university hospitals. We sought to compare online hospital ratings between VA and local affiliated hospitals. This allowed us to compare the patient experience at VA and non-VA hospitals in the same region that often share trainees and physicians. 

We searched Yelp for hospital ratings for the 131 VA hospitals from March 2014 to June 2014 and recorded the Yelp ratings 1 (worst) to 5 (best). Ratings were excluded if the category was specific to a department (eg, cafeteria). We only included hospitals if there were 2 or more online ratings. Therefore, we attempted to identify the local university hospital affiliate of each of the VA hospitals. If there was more than 1 local affiliate, we used the affiliate with the greatest number of online reviews. For 1 VA hospital for which there was no local affiliate, we selected a random hospital within 5 miles that had online reviews. We determined hospital characteristics (bed size, membership in the Council on Teaching Hospitals [COTH], presence of an accredited graduate medical education program, and certification by The Joint Commission [TJC]) using data from the American Hospital Association. 

We determined the mean Yelp rating for each hospital.

Continuous variables of hospital characteristics were compared between VA and non-VA hospitals using 

 tests. Categorical variables of hospital characteristics were compared using χ

 tests. We first compared the mean Yelp ratings of the VA and affiliated hospitals using an unweighted paired 

 test. In a sensitivity analysis, comparisons of mean ratings were weighted by the number of ratings using analysis of variance. The Shapiro-Wilk test did not reject the assumption of normality for the rating distribution (

= .94). We then examined the correlation (clustering) of online ratings for the pairs of VA and non-VA hospitals. Finding negligible clustering (see results for correlation), we used ordinary least squares multivariate regression to compare mean Yelp ratings, adjusting for hospital characteristics (bed size, membership in COTH, presence of an accredited graduate medical education program, and certification by TJC). We repeated the analyses limiting the sample to those with 5 or more, 10 or more, and 15 or more individual ratings. A 2-sided 

 <.05 was considered statistically significant. All analyses were performed with Stata version 11.0 (Stata Corp; College Station, Texas). 

We identified 39 VA hospitals (30%) and 39 non—VA-affiliated hospitals with at least 2 Yelp ratings. Hospital characteristics are displayed in the 

. Non-VA hospitals were larger than VA hospitals. The vast majority of both VA and non-VA hospitals were affiliated with medical schools, although non-VA hospitals were more likely to be a member of COTH. Almost all used TJC for hospital accreditation. VA hospitals with less than 2 Yelp reviews (excluded from the study, n = 92) were smaller and less likely to be academic than those VA hospitals with 2 or more ratings (

The mean patient rating (± standard deviation) for VA hospitals was higher (3.64 ± 1.0) than the rating for affiliated hospitals (3.09 ± 0.75; 

 test-unweighted). The median difference between VA and affiliate ratings (VA — non-VA) was 0.5 (interquartile range, –0.5 to 1.5). After weighting for the number of reviews (analysis of variance), the mean patient ratings were slightly higher for both VA hospitals (3.70 ± 0.74) and affiliated hospitals (3.19 ± 0.54), although the difference remained similar and significant (

= .0025). This difference in online ratings persisted when the sample was limited to dyads of hospitals (VA and non-VA) with more than 2 reviews (

). There was no significant correlation in rating between a VA hospital and its affiliate (r = 0.07; 

= .59). After adjustment for hospital characteristics, the adjusted rating difference (VA-affiliate hospital) was 0.65 (95% confidence interval, 0.18-1.12). 

Patient satisfaction with care is increasingly recognized as important by payers, including CMS. Hospital ratings on the HCAPHS surveys are publically reported

 and used as part of pay-for-performance programs. Patient-initiated reviews have limitations, but a recent study found a strong correlation (r = 0.49; 

<.0001) between Yelp reviews and HCAHPS scores for non-VA US hospitals.

 At the time of the study (2011), 25% of all non-VA US hospitals had a Yelp review, and this percentage increased over time—this percentage is consistent with our ability to find reviews for 30% of VA hospitals. The mean score for all non-VA US hospitals was 3.3 in 2011, slightly higher than the non-VA hospital score in our study. 

Although Yelp reviews describe patient experience, they may also be a marker of patient outcome. Our study was not powered to detect differences in outcome; however, the Bardach study

 found that high Yelp ratings were significantly associated with lower 30-day all-cause mortality following an admission for myocardial infarction or pneumonia and lower 30-day all-cause readmissions following a discharge for myocardial infarction, heart failure, or pneumonia.

Our study is potentially limited by our comparison group, which included local university-affiliated hospitals that often share trainees and faculty with the VA hospital. Although veterans may often choose the local affiliate hospital to receive care, our results may not be applicable to VA comparisons with non-affiliated community hospitals. We excluded 70% of VA hospitals with fewer than 2 reviews and these were smaller and less academic than the included studies sites. A prior study found that smaller hospitals had higher patient satisfaction ratings than larger hospitals although there was no significant impact of teaching status on ratings.

 Differences in patients could explain the results. For example, we cannot know if veterans are more likely than nonveterans to rate all hospitals highly.

We found that VA hospitals have higher patient ratings than their affiliated hospitals, a finding not explained by bed size or teaching status.

 Veterans Affairs Palo Alto Healthcare System (PAH, AZ, NO, AS), Palo Alto, CA; Department of Medicine, Stanford University School of Medicine (PAH, LS)

US Department of Veterans Affairs HSR&D CHF 04-236.

 The hospital where Drs Heidenreich, Zapata, Oliva, and Sahay were employed was one of the study sites. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (PAH, AS); acquisition of data (PAH, AZ); analysis and interpretation of data (PAH, AZ, LS, NO, AS); drafting of the manuscript (PAH, LS); critical revision of the manuscript for important intellectual content (PAH, LS, NO, AS); obtaining funding (PAH); administrative, technical, or logistic support (AZ, NO).

Paul Heidenreich, MD, MS, VA Palo Alto HCS, 111C, 3801 Miranda Ave, Palo Alto, CA 94304. E-mail: paul.heidenreich@va.gov.

1. Department of Veterans Affairs. Expanded access to non-VA care through the Veterans Choice Program [interim final rule]. 

2. Centers for Medicare and Medicaid Services. Hospital Compare website. http://www.medicare.gov/hospitalcompare. Published 2014. Accessed January 23, 2015. 

3. Bardach NS, Asteria-Peñaloza R, Boscardin WJ, Dudley RA. The relationship between commercial website ratings and traditional hospital performance measures in the USA. 

. 2013;22(3):194-202. doi: 10.1136/bmjqs-2012-001360.

4. Ranard BL, Werner RM, Antanavicius T, et al. Yelp reviews of hospital care can supplement and inform traditional surveys of the patient experience of care. 

. 2016;35(4):697-705. doi: 10.1377/hlthaff.2015.1030.

5. Elliott MN, Cohea CW, Lehrman WG, et al. Accelerating improvement and narrowing gaps: trends in patients’ experiences with hospital care reflected in HCAHPS public reporting. 

. 2015;50(6):1850-1867. doi: 10.1111/1475-6773.12305.&ensp;

Articles

The mean online patient rating for Veterans Affairs hospitals was higher (3.70 ± 1.3 out of 5) than the rating for affiliated hospitals (3.19 ± 1.3; P = .003).

Patient Ratings of Veterans Affairs and Affiliated Hospitals

 To determine whether atrial fibrillation (AF) patients with mental health conditions (MHCs) were less likely than AF patients without MHCs to be prescribed warfarin and, if receiving warfarin, to maintain an International Normalized Ratio (INR) within the therapeutic range.

 Detailed chart review of AF patients using a Veterans Health Administration (VHA) facility in 2003.

 For a random sample of 296 AF patients, records identified clinician-diagnosed MHCs (independent variable) and AF-related care in 2003 (dependent variables), receipt of warfarin, INR values below/above key thresholds, and time spent within the therapeutic range (2.0-3.0) or highly out of range. Differences between the MHC and comparison groups were examined using X

 tests and logistic regression controlling for age and comorbidity.

 Among warfarin-eligible AF patients (n = 246), 48.5% of those with MHCs versus 28.9% of those without MHCs were 

 = .004). Among those receiving warfarin and monitored in VHA, highly supratherapeutic INRs were more common in the MHC group; for example, 27.3% versus 1.6% had any INR >5.0 (

 <.001). Differences persisted after adjusting for age and comorbidity.

 MHC patients with AF were less likely than those without MHC to have adequate management of their AF care. Interventions directed at AF patients with MHC may help to optimize their outcomes.

	Mental illness decreased likelihood of warfarin receipt among patients without recorded contraindications, but clinicians did not list mental illness as a rationale for withholding warfarin.

		Among warfarin-treated patients, those with mental illness were more likely to have some International Normalized Ratio (INR) values substantially above the therapeutic range, but some with mental illness consistently maintained therapeutic INRs.

		Identifying subgroups with mental illness who can/cannot safely take warfarin would inform interventions and guidelines.

The Veterans Health Administration (VHA) National Mental Health Strategic Plan has recently called attention to optimizing treatment of coexisting medical conditions in patients with mental illness. Research is beginning to amass suggesting that patients with mental health conditions (MHCs), representing nearly one third of Americans,

 may receive less intensive care for a range of conditions from diabetes to coronary artery disease to chronic pain.

 However, the influence of mental illness on the management of atrial fibrillation (AF) is not yet well understood. This is important because AF is highly prevalent, especially among older adults, and is associated with increased stroke risk.

 An oral anticoagulant, warfarin, reduces AF-related stroke risk from approximately 5% per year to 1% to 2% per year,

 but also has downsides: it increases major bleeding risk, it typically involves a complex and rigorous dosing schedule, and it requires frequent blood monitoring of coagulation status (International Normalized Ratio [INR]). Given warfarin’s narrow therapeutic window, clinicians may have concerns about whether patients with MHC can safely take warfarin as prescribed and follow through with necessary monitoring tests.

Unfortunately, major AF practice guidelines do not currently address whether mental illness should influence anticoagulant therapy decision making.

 This is concerning, since there are potential adverse consequences of prescribing warfarin for a patient who cannot use it safely: excess major bleeding if overdosed or thromboembolic events if underdosed. Likewise, there are potential adverse consequences of not prescribing warfarin for a patient who could take it safely, such as unnecessarily increased stroke risk.

Therefore, a detailed medical chart review methodology was used at 1 VHA facility to determine:

• Whether VHA AF patients with MHCs were less likely than those without MHCs to receive warfarin, even in the absence of contraindications.

• Whether, among those treated with warfarin and monitored in VHA, those with MHCs were more likely than those without MHCs to have INRs falling outside the standard therapeutic range.

Patients were drawn from a national database of AF/atrial flutter cases developed for another ongoing study, where AF was identified by the presence of

 International Classification of Diseases, Ninth Revision (ICD-9)

 code 427.3x in fiscal year 2001-2002 VHA or Medicare outpatient or inpatient administrative data. Patients with AF/atrial flutter were included regardless of whether or not the arrhythmia was thought to be paroxysmal, and regardless of duration of the diagnosis (except for transient AF; see below). From the 2490 members of this database who received more than 50% of their VHA outpatient care in both 2002 and 2003 at a suburban, West Coast VHA facility, a random sample of 450 was selected for detailed chart review of electronic medical records. After refining the standardized abstraction protocol on 50 cases not included in the study, a trained senior medical student identified information on variables of interest and recorded them on a standardized abstraction form. Ongoing quality control occurred at weekly meetings of the research team to ensure consistency in coding.

 Administrative data supplemented chart data for some variables.

Excluded were 6 patients with inaccurate chart record identifiers, 33 patients who died prior to the end of 2003, 17 patients with transient AF (occurring only during an acute hospitalization, or within 30 days of cardiac surgery or cardiac catheterization), and 98 with no VHA chart evidence of AF based on data from available electrocardiograms, patient problem list, or outpatient provider diagnoses from progress notes. (Among this latter subgroup, 75 were identified as having AF by Medicare

 codes only; likely their AF was managed by a non-VHA provider, and thus the diagnosis did not appear in VHA files.) A total of 296 AF cases (full cohort) were thus included in the study. Of these, 51 patients had evidence of 1 or more potential contraindications to warfarin: history of nonadherence, hemorrhage, falls, dementia, advanced liver disease, or malignancy.

 Among the remaining 246 patients, all of whom were apparently eligible to receive warfarin (warfarin-eligible subcohort), 161 (65.4%) received warfarin within or outside of VHA. Among them, 84 received their INR monitoring in the VHA Anticoagulation Clinic and had reliable INR data available, such that warfarin effect could be assessed; they constituted the warfarin-treated-VHA-monitored subcohort. This study was approved by the Institutional Review Board at Stanford University, Stanford, California.

The primary independent variable was MHC. MHC was defined as (1) 1 or more provider diagnosis of substance use disorder, depressive disorder, anxiety disorder, psychotic disorder, dysfunctional personality disorder, or bipolar disorder (in progress notes or problem lists) during 2001-2003; or (2) receipt of psychopharmacologic agents including antidepressants, antipsychotics, or mood stabilizers (based on the VHA medication log or based on a progress note referring to a non-VHA prescription) during 2001-2003.

Dependent variables included (1) receipt of warfarin in 2003 (yes/no), based upon the chart medication log or a progress note referring to a warfarin prescription by a non-VHA provider; (2) presence of at least 1 subtherapeutic INR in 2003 (<1.5, <2.0) based on laboratory result file data; and (3) presence of at least 1 supratherapeutic INR in 2003 (>3.0, >3.5, >4.0, >4.5, >5.0). A standard procedure was used to calculate time in therapeutic range (ie, the percentage of days in 2003 during which the INR fell within the 2.0 to 3.0 range, linearly interpolating INR values for days on which an actual value was not available).

 Time highly out of therapeutic range was calculated as percentage of days in 2003 during which the INR was less than 1.5 or greater than 4.0.

Other variables included known stroke risk factors for AF patients

: age 75 years or older, congestive heart failure, hypertension, diabetes, and prior nonhemorrhagic stroke or transient ischemic attack, based on 2001-2002 VHA outpatient or inpatient administrative data. Comorbidity was identified from the medical component of the Comorbidity Index (a count of medical conditions common in veterans and used for ambulatory care case-mix adjustment), using VHA administrative data.

 A count of primary care visits in 2003 came from clinic type codes in VHA administrative files. Patients were considered to have received only VHA care if they lacked any forms of outside insurance, based on the medical record, and if there was no mention of receipt of outside medical care or medications, based on primary care and cardiology clinic progress notes.

Among the full cohort, the warfarin-eligible subcohort, and the warfarin-treated-VHA-monitored subcohort, the following characteristics were examined for patients with and without MHC: demographics, stroke risk factors, Comorbidity Index, and utilization.

Among warfarin-eligible patients, the next analyses compared the proportions of those with and without MHC who were receiving warfarin, using a X

 test. In this group, multivariable logistic regression was performed to examine receipt of warfarin as a function of the presence of MHC, first unadjusted, then controlling for age, and then controlling for age and Comorbidity Index.

Next, among warfarin-treated-VHA-monitored patients, proportions were sequentially compared for those with and without MHC who had at least 1 low INR (<2.0, <1.5) and for those with and without MHC who had at least 1 high INR (>3.0, >3.5, >4.0, >4.5, >5.0), using a X

 test. Fisher’s exact test was used to calculate 

 values in instances of small cell size. Then logistic regression was used to examine high or low INR as a function of MHC, first unadjusted, then controlling for age, and then controlling for age and Comorbidity Index.

Among warfarin-treated-VHA-monitored patients, patients with and without MHC were compared with respect to mean time in therapeutic range and to mean time highly out of therapeutic range, using 

Sensitivity analyses tested the stability of findings. Because of a concern that some VHA patients may receive their AF care from non-VHA providers, sensitivity analysis 1 examined the effect of restricting analyses to patients who used VHA as their exclusive source of care (ie, for whom all AF care could be reliably captured). Because of the possibility that some patients with a single recording of an AF diagnosis could have been miscoded as being an AF case, or could have had transient AF (despite efforts to exclude transient AF cases), sensitivity analysis 2 examined the effect of restricting analyses to patients who had at least 2 instances of an AF diagnosis from electrocardiograms or primary care/cardiology clinic notes.

Among the full cohort (n = 296), 29.7% had at least 1 diagnosed MHC: substance abuse 7.8%, depression 18.9%, anxiety 15.2%, psychosis 3.0%, dysfunctional personality 1.0%, and mania 0.7%. Patients with MHC tended to be younger than those without MHC, yet tended to have greater medical comorbidity, including stroke risk factors (

). Among the 17.2% of patients with a listed contraindication to warfarin, the most common contraindications cited were dementia (7.8%), history of falls (2.7%), and history of hemorrhage (3.7%). Of note, no records mentioned MHC as a contraindication to warfarin.

In the warfarin-eligible subcohort, 48.5% of those with MHC versus 28.9% of those without MHC were not treated with warfarin (

); even after adjustment for age and comorbidity, those with MHC were significantly less likely to receive warfarin (

). The frequency of not being treated with warfarin varied by specific MHC, as follows: substance abuse 66.7%, depression 50.0%, anxiety 47.4%, psychosis 25.0%, and personality disorder 50.0%. Note that among the warfarin-eligible subcohort (n = 246), 19.7% of patients with MHC and 15.6% of patients without MHC had a CHADS2 (congestive heart failure, hypertension, age 75 years or older, diabetes mellitus, previous stroke or transient ischemic attack) score below 2, a level at which alternatives to warfarin may be appropriate.

Among the warfarin-treated-VHA-monitored subcohort (n = 84), patients with MHC were more likely than patients without MHC to have high INR values, both before and after adjustment for age. This would put them at increased risk for hemorrhage. There was a trend toward the same effect after additionally adjusting for Comorbidity Index. Differences between patients with versus without MHC were even more pronounced at higher INR thresholds. There was also a nonsignificant trend toward higher rates of subtherapeutic INR (<1.5) among patients with MHC.

Among the warfarin-treated-VHA-monitored subcohort, mean time spent in the 2.0 to 3.0 therapeutic range was lower for the MHC group than for the group without MHC: 56.8 (standard deviation [SD] 16.9) versus 65.9 (SD 18.2) percent of days (

 = .04). Mean time spent highly out of range was higher for the MHC group than for the group without MHC: 9.0 (SD 9.8) versus 3.4 (SD 7.1) percent of days (

In sensitivity analysis 1 (restricting to patients who received all care in VHA), the magnitude and direction of unadjusted differences were similar to those in the main analyses. The same was true in sensitivity analysis 2, restricting to patients with 2 or more AF diagnoses (

Warfarin-eligible AF patients with MHC were substantially less likely than those without MHC to receive warfarin, even after controlling for age and comorbidity. This is of potential concern, since stroke risk factors were greater in AF patients with MHC than in those without MHC. When prescribed warfarin, patients with MHC were substantially more likely than those without MHC to have INRs in the highly supratherapeutic range, with 27% having at least 1 INR value above 5.0. There was also a trend for them to have greater risk of having INRs in the highly subtherapeutic range. As a result, these patients could have increased risk of either thromboembolism (in the setting of low INR) or hemorrhage (in the setting of high INR).

The differences in warfarin receipt observed here are consistent with emerging literature documenting lower-intensity care provided to patients with MHC for a range of medical conditions,

 and could reflect several potential mechanisms. Patient preferences could be important,

but have not been well characterized with respect to MHC. Mental illness can influence physicians’ perceptions of patients,

 and physician expectations of patients—founded or not&mdash;may affect management.

 Alternatively, physicians’ threshold for concluding that falls, dementia, or nonadherence (known to affect prescribing behavior

) should be counted as a warfarin contraindication may be lower for patients with MHCs. Mental illness may also affect communication between physician and patient, or may cause patients to be labeled as “difficult.”

 While sample sizes were small for specific MHCs, this study did find preliminary evidence that warfarin receipt varied by specific MHC, and that rates were particularly low for patients with substance use disorders, suggesting that risky behaviors could factor into clinical decision making. Other possibilities are that providers may simply assume that patients will be nonadherent,

 or may worry about drug interactions with psychiatric medications, some of which can increase bleeding risk.

 Patient and provider-level contributors to this effect merit more inquiry.

While these findings reflect care provided at a single site and thus will require replication, they raise several issues of serious clinical relevance. Findings shed light on actual physician behavior: in the absence of explicit national guidelines to shape their decisions, clinicians are prescribing warfarin substantially less often to AF patients with MHC than to AF patients without MHC. This suggests that the presence of MHC may be influencing their decisions, even though no physicians in this study listed the presence of MHC as the rationale for withholding warfarin. But is this concern warranted? In one sense, the answer is yes: patients with MHC were more likely to have supratherapeutic INR values, which may explain why another study (which focused on warfarin-treated Medicaid patients) found higher rates of bleeds in those with MHC than in those without MHC.

 In another sense, the answer is no: nearly 20% of patients with MHC were able to maintain an INR level below 3.0 throughout the full year of surveillance. Thus, for at least a subset of AF patients with MHC, it may be possible to administer warfarin safely. Perhaps, for example, different types of MHC have different effects upon patients’ ability to take warfarin safely. More work with larger samples is needed to help clinicians distinguish which patients with MHC are likely to have adverse effects, and which are likely to do well on warfarin.

Some have suggested that perhaps newer agents like dabigatran that do not require monitoring

 may be better options for patients with MHC. This may be true: perhaps the simpler dosing regimen that dabigatran provides would facilitate adherence, which can be an issue for patients with MHC.

 The reverse may also be true: perhaps patients with MHC do well in anticoagulation clinics that provide close, mandatory follow-up and that allow clinicians to monitor them. In contrast, medications like dabigatran, which lack an easy laboratory test for monitoring and currently lack a reversal agent for use if taken in overdose or with bleeding, may not offer this advantage. Further research will be needed to understand the pros and cons of dabigatran in patients with MHC.

This study was limited by possible underascertainment of medical care received (ie, non-VHA care), although the stability of findings in sensitivity analyses limited to patients who use VHA as their exclusive source of care was reassuring. This study did not distinguish between AF and atrial flutter, nor between continuous and paroxysmal AF. Also, it is unknown how many patients had ablation or surgical procedures for their AF; however, at the time, within the facility studied, ablation and appendage removal were not being offered and surgical procedures were very rare and limited. Contraindications to anticoagulation may also be underdocumented in medical records,

 although from a medicolegal standpoint, clinicians have an incentive to document why they are withholding warfarin. Results may not be generalizable to women or other geographic or healthcare settings.

This study also has several strengths. The mixed-modality approach made it possible to take advantage of information available in administrative data, such as all INR values across an entire year of surveillance. Supplementing this source with medical record review data facilitated identification of factors difficult to elicit from administrative records, such as clinicians’ decision-making logic (eg, the reasons they did not start warfarin) and information about receipt of care outside of VHA (eg, evidence of receipt of care in general and receipt of warfarin in particular from non-VHA sources).

Atrial fibrillation patients with mental illness are less likely to receive warfarin anticoagulation; those who do receive warfarin have excess risk of over-anticoagulation.

Paul A. Heidenreich, MD, MS

Articles

Patient Ratings of Veterans Affairs and Affiliated Hospitals

Mental Illness and Warfarin Use in Atrial Fibrillation