Design and setting

The University of Pittsburgh/University of Pittsburgh Medical Center, Penn State College of Medicine, Temple University Hospital and Johns Hopkins University (PaTH) Clinical Data Research Network uses clinical data from electronic medical records (EMRs) combined with PROs.14 The PaTH AF is a cohort study at five academic medical centres.14 The PaTH AF has two components: (1) a clinical database of over 95 302 individuals identified through EMRs; this dataset is thus limited to deidentified information from EMRs, and (2) a cohort of 953 consented individuals who were recruited from the larger clinical database. Patients provided electronic informed consent to: (1) complete questionnaires every 6 months on their PROs and (2) allow researchers to extract data from their EMRs and link these data to their PROs.

Inclusion and exclusion criteria

Inclusion criteria for the PaTH AF clinical database included: diagnosis of AF by ICD-9-CM (International Classification of Diseases, 9th Revision, Clinical Modication) and ICD-10-CM (International Classification of Diseases, 10th Revision, Clinical Modification) codes or electrocardiographic reading, three outpatient, non-emergency and non-inpatient visits since 1 January 2011, age ≥18 years and ability to speak, read and understand English. Individuals were excluded if they received the diagnosis of AF within a month of cardiac or abdominal surgery, or a thyroid-related diagnosis, and 12 months before or after prescription for methimazole or propylthioracil. All patients meeting the criteria at each institution were in PaTH the clinical database.

Recruitment, screening and data collection procedures

Individuals identified in the clinical database were recruited through in-person, email, phone, patient portal messaging and post mail beginning in August 2015 (online supplementary figure A1).15 The consented participants were compared on key characteristics to the individuals in the clinical database (online supplementary table A1). Participants recruited through email and post mail completed surveys through REDCap (Vanderbilt University, Tennessee), and participants recruited through the patient portal completed the surveys in MyChart (EPIC software, Wisconsin). All surveys could be completed by the patient online. Participants completed questionnaires on PROs at time of enrolment and every 6 months following enrolment. Healthcare utilisation, procedures, medications, diagnoses and body mass index (BMI) are collected through EMRs. This analysis consists of participants’ first 3 PRO surveys (time of enrolment, 6 months and 1 year).

Patient-reported outcomes

Symptoms and quality of life were measured using the Atrial Fibrillation Effect on QualiTy of Life (AFEQT).16 The AFEQT is a disease-specific quality of life instrument consisting of four subscales: symptoms, treatment concerns, daily activities and treatment satisfaction. The subscales for symptoms (four items), treatment concerns (six items) and daily activities (eight items) are added together to create an AF-related quality of life score (18 items). The AFEQT has high internal consistency (Cronbach’s α: >0.8) and has been validated in a large, community-based cohort.16 In this sample, the AFEQT had a Cronbach’s α of 0.93. The symptoms subscale consists of four items rated on a 7-point Likert scale and has a previously reported Cronbach’s α of 0.516 and a Cronbach’s α of 0.81 in this sample. The symptoms subscale score formed by calculating the mean score of the questions answered. The AFEQT total scale scores range from 0 to 100, with 0 representing the worst possible quality of life.

Emotional and functional status were measured using the PROMIS (Patient-Reported Outcomes Measurement Information System)-29 emotional status (anxiety and depression) and functional status domains.17 The emotional status domain consists of four questions related to anxiety and four questions related to depression rated on a five-point Likert-type scale. The functional status domain consists of four items related to functional limitations rated on a five-point Likert-type scales. High internal consistency (Cronbach’s α: >0.8) for both the emotional and function status domains has been previously reported,17 and was similarly found in this sample. The raw scores of the anxiety, depression and functional status domains were converted to a scaled T-score that ranges from 41 to 79.4 for depression, from 40.3 to 81.6 for anxiety and from 22.9 to 56.9 for function using PROMIS scoring guidelines.18 The scaled T-score is created based on PROMIS scores’ mean of 50 and SD of 10 in a generalised, referent population.

AF therapies and sex

Rate control (rate control medication), rhythm control (rhythm control medication, cardioversion and/or ablation) and sex were the key independent variables. Each was analysed as a binary variable. Patients were classified as receiving rate control medication or rhythm control medication if they were prescribed a medication in an outpatient setting in the respective category within 1 year prior to enrolment. One year was used as a cut-off point since prescriptions are renewed on an annual basis. Rate and rhythm control medications were categorised based on practice guidelines.1 Ablation was classified based on provider billing code CPT 93656 within the past 3 years. Cardioversion was classified based on provider billing codes CPT 92960, 92961 or ICD-9 99.61 within the past 3 years. Three years was used as a cut-off point since ablations and cardioversions are more commonly used as a one-time procedure to correct rhythm or limit symptom severity.1 Sex was determined through EMR documentation.

Other individual and health characteristics

Other independent variables were assessed as potential confounders. Data for age, education (>college (ref.), college, some college and ≤high school) and race (white (ref.), black and other) were collected in the baseline questionnaire. Comorbidities and BMI (normal weight (ref.), underweight, overweight and obese) were determined through EMR data. Comorbidities were measured using the grouped Charlson comorbidities index score, which is calculated using the Deyo version.19 The grouped Charlson comorbidities index score rates comorbidities on a 0–2 scale. Dummy variables for study site were used.

Statistical methods

We used the statistical software STATA V.15.0 for all analyses. Patterns of missing data were assessed for each variable included in the models and, for variables with greater than 5% missing, missing data were imputed using multiple imputation.20 We created 100 imputations for each PRO since 13% of these measures were missing. When we compared results using completed cases (n=829) only against results using multiple imputation (n=953), we found that inferences did not differ across the models. We conducted a power analysis with an alpha of 0.05 and determined that we had power >0.80, thus attention to effect sizes associated with significant results was important.

Propensity-score matching was used to address potential confounding of the association between AF therapies and PROs. Nearest neighbour matching (1:10) with replacement was conducted separately for individuals solely on rate control medications and individuals on rhythm control strategies. Analyses were restricted to individuals who could be matched to another participant with a similar probability (calliper ≤0.04) of receiving the therapy to reduce the possibility of model extrapolations and the biasing effects of model misspecification.21 The individuals on rhythm control strategy (n=325) were matched to 525 ‘control’ individuals, and individuals on rate control medication only (n=144) were matched to 708 ‘control’ individuals. The propensity score models for rate control and rhythm control strategies reduced bias on key variables (online table A2). Hypothesised associations between rate control and rhythm control strategies and each PRO were tested separately using linear regression models among propensity score matched subsamples. All models adjusted for individual characteristics (sex, age and education level) and health characteristics (BMI and comorbidities). Regression coefficients were not standardised in order for the PROs to be interpreted in their unit of measurement.

A subset of individuals underwent a rhythm control strategy including ablation, cardioversion or rhythm control medication (n=111) or only a rate control medication (n=26) after completing their baseline PROs survey, which allowed for comparisons in PROs change following therapy. We analysed the mean change in symptoms of AF, depression and anxiety, and functional status in bivariate regression models to determine the effect of rhythm and rate control medication for this subset of individuals.

Sample characteristics

Table 1 depicts demographic data specific to participants stratified by recent rate and rhythm control therapies. Our study population (n=953) were 65% male (n=616), 93% white (n=890) and 72 (±10) years old on average. Thirty-seven per cent of participants had less than a college degree (n=330). A comorbidity index of 2 was present in 39% (n=369) of the sample. The common comorbidities were hypertension (71%), heart failure (28%) and diabetes (22%). The majority of the population was overweight (35%) or obese (44%). History of ablation was present in 7% (n=70), 21% had a cardioversion (n=190), 56% had a history of being on rate control medication (n=538) and 58% had a history of being on rhythm control medication (n=554). Rhythm control medication had been prescribed to 19% (n=182) of the sample in the past year, and rate control had been prescribed to 16% (n=158) of the sample in the past year.

Our sample reported poorer average functional status (47.3), and lower symptoms of anxiety (48.9) and depression (46.9) than the general population mean. Participants reported an average of 2.21 on the AF symptoms subscale, which can be interpreted as the participants were on average ‘a little bothered’ by their AF symptoms. On average, the AFEQT total score was 62.72 on a scale of 0–100, with 0 being the lowest AF-related quality of life.

Rate control strategies matched sample

Among the sample matched on rate control medication prescription, rate control medication was associated with significantly poorer functional status (−3.69, 95% CI −5.27 to –2.12) and AF-related quality of life (−7.22, 95% CI −11.51 to –2.92) (table 2, figure 1). Rate control medications were not associated with anxiety, AF symptoms and depression. Among the subset of individuals (n=25) who had a documented rate control medication prescription only after the baseline PROs, the mean change in PROs was not significantly different from individuals with no documented rate or rhythm control strategies (see online appendix table A4).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The association of therapies and key patient characteristics with atrial fibrillation-related quality of life in propensity score matched adjust models. Notes: results presented as β, 95% CI. The reference groups for sex, education level, comorbidity index and weight were male sex, greater than college, comorbidity index of 0 and normal weight. *Comorbidity  index.

Women were more likely to report poorer functional status, higher symptoms of anxiety, depression and AF and AF-related quality of life. Older age was associated with poorer functional status, lower symptoms of anxiety, depression and AF symptoms and high AF-related quality of life. Compared with individuals with greater than a college education, individuals with less education reported poorer functional status. Individuals with some college reported higher symptoms of AF and anxiety and poor AF-related quality of life compared with individuals with greater than a college education.

Rhythm control strategies matched sample

Among the sample matched on history of rhythm control strategies, rhythm control strategies were not associated with PROs (table 3). Among the subset of individuals (n=111) who had a documented rhythm control medication prescription only after the baseline PROs, the mean change in PROs was not significantly different from individuals with no documented rate or rhythm control strategies (see online appendix).

Women reported higher symptoms of AF, anxiety and depression and poorer functional status and AF-related quality of life. Individuals with a college education, some college and less than college had poor functional status compared with individuals with greater than a college education. Individuals with some college education had higher symptoms of AF and anxiety and poorer AF-related quality of life compared with individuals with greater than a college education. Older age was associated with poorer functional status, lower symptoms of anxiety, depression and AF and higher AF-related quality of life.

Clinical implications

AF symptoms are a chief driver of clinical decision making for AF management.1 AF symptoms, emotional status, quality of life and functional status are all linked to each other.2 3 5 22 Yet in this study, we found that AF therapies received were not associated with more favourable PROs. Women in this study reported higher AF symptom severity, and they also reported higher anxiety and depression and poorer functional status. Identifying and addressing root causes for individuals’ higher anxiety and depression and poorer functional status may improve AF-related PROs. Risk factors for AF, including hypertension, diabetes, obstructive sleep apnoea, obesity, poor diet, smoking and alcohol use,30 may contribute to comparatively poorer PROs. Managing these conditions as part of a comprehensive treatment approach may improve functional and emotional status and quality of life, which may, in turn, improve AF symptoms.

Limitations

There are important limitations to this study that must be acknowledged. The differences in PROs are minimally clinically significant: AFEQT requires a 12-point change for a clinically significant change16 and PROMIS similarly requires a change by at least five points for clinical significance.18 Recruitment was based on EMR documentation of AF, and thus, biases exist in patient selection. Men, people with fewer comorbidities and younger adults were more likely to join this study. Both this study sample and the larger PaTH AF clinical database individuals were majority white, which may be reflective of the population in treatment for AF. There is a lower prevalence of this disease among black and Hispanic populations that adds to the challenge of enrolling a racially diverse group of patients.

Relying on EMR data for information on therapies is a significant weakness. Whether the therapies were successful (ie, if rate and rhythm were controlled) is not captured. EMR data are subject to biases that exist from institution to institution and provider to provider and are limited to the care the individuals received at the institutions involved in the study. We chose conservative time windows for therapies in recognition of EMR data limitations. We conducted sensitivity analyses with alternate time windows for medications (history of prescription vs limiting to prescriptions 1 year prior) and ablations (limited to a 1-year window vs a 3-year window), and our findings were unchanged. Agreement between patient-reported medication use and documented prescription of medication was assessed in a subset of patients, and high agreement was found using medications prescribed in the past year (κ >0.80).