We performed a prospective cohort study that was nested in the intensive self-care intervention arm of a randomized clinical trial that compared different levels of HF self-care training [7]. Details of this trial have been described previously and will be summarized here. Patients in this study had either diastolic or systolic HF with New York Heart Association (NYHA) class II-IV symptoms within the prior 6 months, adequate cognitive function, were on a loop diuretic, and were fluent in either English or Spanish. Patients were recruited between 2006 and 2009 from Internal Medicine and Cardiology clinics at the University of North Carolina (UNC) at Chapel Hill, Northwestern University, Olive View – UCLA Medical Center, and UCSF – San Francisco General Hospital. Institutional Review Boards (IRB) from UNC, Northwestern University, Olive View-UCLA, and UCSF approved the protocol and all patients provided informed consent. The investigation conformed to the principles outlined in the Declaration of Helsinki.

All patients in the intensive self-care intervention arm received an in-person 40-minute education session followed by a series of educator calls (median of 15) over 1 year. The education session and calls reinforced the importance of daily weight monitoring, taking proper diuretic doses, medication adherence, salt reduction, and exercise. Most patients (72% of intervention arm patients) were taught how to adjust their diuretic doses based on their daily weights. Among the four educators who delivered the intervention, two were registered dieticians with experience counseling patients in clinical settings; the other two had bachelor’s degrees and experience working as health educators. The educators convened for a one-day training prior to enrollment and participated in weekly calls with an investigator to develop the educational protocol and ensure similar education delivery across sites.

All patients were provided a digital bathroom scale and a diary in which they were to record their daily weights. Patients were instructed to mail weight diaries back to their educator on a monthly basis in self-addressed, stamped envelopes. The educator reviewed the diaries to assess how well a patient was following the protocol and incorporated that information in their ongoing interactions with that patient. At the conclusion of the study period (12 months after enrollment), trained study personnel collected the diaries from each site and entered them into a central database for analysis.

The initial number of patients in the intensive self-care training arm was 303. For this analysis, we made exclusions for submitting no diary data during the trial (n = 67) or for a missing response to the self-reported weighing adherence measure from the 12 month survey (n = 20 additional exclusions), leaving a sample of 216 patients.

A trained research assistant (RA) at each site administered baseline questionnaires and collected information about race/ethnicity, income, and insurance status. Socioeconomic status (SES) was measured with the MacArthur Scale of Subjective Social Status; this measure asks individuals to indicate on which ladder rung they feel that they stand relative to other people in the United States with regard to money, education, and jobs (range 0–9) [8]. HF severity was assessed using a single item with a series of structured statements to categorize patients by New York Heart Association (NYHA) class. HF-related quality of life was assessed using the Improving Chronic Illness Care Evaluation (ICICE) HF Symptom Scale (HFSS); [9] HFSS scores were transformed to a 100-point scale with 100 representing the least symptoms/best health. Literacy was measured with a Short Test of Functional health Literacy in Adults (s-TOFHLA) and categorized into adequate literacy (≥ 23 out of 36) or inadequate/marginal literacy (< 23 out of 36) [10]. The RA abstracted medical records to collect additional demographic and clinical data.

The primary end-point for this study was HF-related hospitalization. The University of North Carolina Survey Research Unit conducted telephone surveys at 1, 6, and 12 months during which patients were asked about any hospitalizations or other events that occurred in the previous time period. Upon identification of a hospitalization, admission and discharge summaries were requested from the hospital; these records were reviewed by one of three adjudication committee members masked to intervention status to determine if the hospitalization was HF-related. Adjudication committee members used a study protocol and their clinical judgment to assess whether HF was present at admission and whether HF was an important contributing factor to the hospitalization. If the first assessor concluded that a hospitalization was definitely HF-related or definitely not HF-related, no further assessment was done. For all other hospitalizations, another assessor conducted a second assessment. If the first two assessors disagreed, a third assessor reviewed the hospitalization and the case was discussed by the full committee to resolve differences.

Weight monitoring adherence was measured using two methods: self-reported frequency of weighing and diary-recorded weights. Self-reported weight monitoring adherence was collected at baseline and during telephone interviews using the following question: “How often do you weigh yourself?” Patients who reported weighing at least once a day (“every day” or “twice a day or more”) were considered optimally adherent; those who reported weighing less frequently (“several times a week”, “once every week or two”, or “less frequently than every 1 to 2 weeks”) were considered sub-optimally adherent. This threshold was chosen because weighing “every day” indicates a stricter adherence level that is closer to our ≥80% diary threshold than “several days a week”, which could represent a wide range of adherence. Our main predictor variable in the self-reported weight monitoring adherence analysis was collected following completion of the full intervention at the 12-month interview. Diary-recorded weight monitoring adherence was determined from weight diaries and dichotomized into ≥80% and <80% days of adherence to weight monitoring over a patient’s days in the trial to reflect optimal and sub-optimal adherence, respectively. The ≥80% threshold was chosen because this threshold is commonly used to define optimal medication adherence, [11‐13] and because we used this threshold to define weight monitoring adherence in a prior publication [6]. We also performed sensitivity analyses to evaluate how different thresholds of weight monitoring adherence were associated with HF-related hospitalizations.

We compared demographic and clinical data by adherence groups for both self-reported and diary-recorded data for descriptive purposes with a chi-square test for dichotomous variables and Student’s t-test for continuous variables. A Fisher’s exact test was used to compare dichotomous variables with 5 or fewer individuals in 25% of categories. A Kolmogorov-Smirnov test was used to assess for normal distribution of continuous variables; for continuous variables found to have a non-normal distribution, a Wilcoxon-Mann–Whitney test was used to compare groups.

We compared the incidence rates of HF hospitalizations between (1) self-reported optimal versus sub-optimal adherence groups and (2) diary-recorded optimal versus sub-optimal adherence groups using negative binomial regression. In one negative binomial regression model we adjusted only for site; in the fully adjusted negative binomial regression model, we included demographic and clinical factors that were either statistically significantly different between groups in bivariate analyses (i.e., site, age, race/ethnicity, SES, HF QOL, previous MI/angina, self-reported weighing at baseline, and systolic dysfunction) or have been previously associated with HF-related hospitalizations (i.e., gender) [14]. Standard errors were adjusted for clustering by site. Of note, we did not include total educator calls in the final model because educator calls were likely to have been part of the causal pathway in the association between weight monitoring adherence and HF-related hospitalizations.

We performed sensitivity analyses in which we used different adherence thresholds for diary-recorded weight monitoring adherence (i.e., ≥90%, ≥70%, and ≥60% thresholds). We considered a two-sided p value of <0.05 statistically significant. All data analyses were performed using Stata 12.0 (College Station, TX).

Among 216 patients the mean age was 62.2 years, about half were male, and 40% were African American (Table 1). Nearly half of patients reported an income < $15,000/year, 34% had inadequate literacy, and approximately half had NYHA Class II HF. Of interest, only 28% of patients reported weighing daily or more frequently at baseline. The 87 patients excluded from this analysis because they either submitted no diary data during the trial (n = 67) or a were missing a 12 month self-reported weighing adherence measure (n = 20) were more likely to be younger (58.4 years), male (63%), have NYHA Class III or IV HF (42%), and were less likely to have a prior MI or angina (25%).

At 12 months, 85% of patients completing a survey (184 of 216) self-reported weighing themselves at least daily (Table 1). Those self-reporting weighing daily or more were more likely to be White, Non-Hispanic (39% versus 28% reporting less than daily, p 0.024), have a higher SES (score of 5 versus 4, p 0.026) , report previous MI or angina (41% versus 22%, p 0.037), systolic dysfunction (61% versus 34%, p 0.005), and on average completed one more telephone call with the clinical educator than patients who reported weighing less than daily (16 calls versus 15 calls, p <0.001, Table 1).

Over 12 months, 50% of patients (107 of 216) recorded and returned ≥80% of daily weights. Those who recorded ≥80% of weights were more likely to be older (64.3 versus 60.1 years for <80% weighing adherence, p 0.024), report a higher HF quality of life (score of 64 versus 57 of 100, p 0.014), report previous MI or angina (46% versus 31%, p 0.027), and on average completed two more telephone calls with the clinical educator than patients who recorded <80% of daily weights (17 versus 15 calls, p <0.001, Table 1). Those who were taught weight-based diuretic self-adjustment were distributed evenly among all groups.

Among 216 patients, we identified 50 HF-related hospitalizations over 12 months; 20 patients had 1 hospitalization, 8 patients had 2 hospitalizations, and 4 patients had 3 or more HF-related hospitalizations. Patients with optimal self-reported weight monitoring adherence at 12 months had 0.26 events per person-year (47 events in 184 patients) and those self-reporting sub-optimal adherence had 0.09 events per person-year (3 events in 32 patients). Patients with ≥80% diary-recorded weight monitoring adherence had 0.14 events per person-year (15 events in 107 patients) and those with <80% diary-recorded weight monitoring adherence had 0.32 events per person-year (35 events in 109 patients).

Table 2 shows incidence rate ratios (IRR) for HF hospitalizations both for self-reported and diary-recorded weight monitoring adherence. Ten patients were missing one or more covariate and were excluded from the adjusted models (adjusted model N = 206). In those with optimal self-reported weight monitoring adherence, the IRR for HF hospitalization 1.99 (95% CI 0.69-5.72) adjusted only for study site and 1.34 (95% CI 0.24-7.32) in a fully adjusted model; neither result was statistically significant. In those with diary-recorded weight monitoring adherence of ≥80%, we found a statistically significant lower incidence of HF hospitalization, both in a model adjusted only for study site (IRR 0.32, 95% CI 0.15-0.70) and in a fully adjusted model (IRR 0.37, 95% CI 0.18-0.75).

When we changed the diary-recorded optimal adherence threshold to ≥60%, ≥70%, and ≥90%, the magnitude of effect strengthened with each step up in adherence (Table 3). Thresholds for adherence that ranged down to ≥60% were associated with a lower risk for HF-related hospitalization that was statistically significant. Figure 1 demonstrates the association between HF-related hospitalization and diary-recorded adherence as a continuous measure over 12 months; of note, confidence intervals widened considerably with diary-recorded adherence less than 60%, which was related to smaller sample sizes. To address potential selection bias that may have been introduced by excluding patients who never submitted a diary, we performed sensitivity analyses including this group and found similar IRRs for HF-related hospitalizations when we adjusted only for study site, both for the self-reported recall measure (IRR 1.92, 95% CI 0.80-4.62) and for the diary-recorded measure of adherence (IRR 0.24, 95% CI 0.14-0.41).

With regard to study limitations, our self-reported adherence question did not specify the duration of time for which patients were to report weighing behavior, which likely increased recall bias in responses. In addition, given that 85% of patients self-reported weighing at least daily at 12 months, patient responses to this question may not have accurately reflected behavior, potentially due to social desirability or other factors. It is important to note that the surveys were delivered by masked outcome assessors. The smaller sample size and very low rate of HF-related hospitalization in the group that self-reported sub-optimal adherence contributed to less precision in our results and lack of statistically significant findings.

In addition, our measure of diary-recorded weight monitoring averaged adherence over the entire follow-up period. As such, it is possible that adherence directly preceding HF hospitalizations differed from overall adherence; however, in a prior case–control study nested in this same study population, we found that diary-recorded weight monitoring adherence (≥80%) in the week preceding HF emergency department visits or hospitalizations was similarly protective [6].

Our comparison between self-reported and diary-recorded adherence measures was not a direct comparison between randomized groups of individuals, but instead a comparison of how different measures of adherence were associated with clinical outcomes in the same group of individuals. We noted baseline differences between groups who were optimally adherent and sub-optimally adherent to both self-reported and diary-recorded measures; thus selection bias or unmeasured confounding may have affected our results. However, we attempted to mitigate such differences by adjusting for these variables in our analysis; in addition, we found comparable results to our main analysis when we performed a sensitivity analysis in which we included patients who never submitted a weight diary. The generalizability of our findings may be limited by our younger patients with less advanced HF than in other studies of HF self-care [1, 26]. Strengths of this study include a diverse, well-characterized sample drawn from a multi-site trial and rigorous outcome assessment.