We designed a population-based retrospective cohort study using the resources of a project published in clinical trials database (NCT00792402), which evaluated the impact of a computerized clinical guideline on CHF in Catalonia (population of 7 210 50821). Briefly, in this project, a pragmatic, nonequivalent, controlled, before-and-after quasi-experimental study was performed using a population-based approach. Patients from 2 regions (urban and rural) were included. This project was established to evaluate a complex intervention in the urban setting where primary care practices (PCPs) were randomized. Half the PCPs followed usual care management plus a computerized clinical guideline on CHF and half underwent a disease management intervention.

For the purpose of this study, data from both regions were combined. Despite urbanization differences, both regions shared the same organisational features22 (Table 1). Ethical approval for this study was obtained from the Health Care Ethics Committee of the Institut d’Investigació en Atenció Primària Jordi Gol.

We included incident patients attending PCPs from 2005 to 2007. Patients aged more than 30 years old with a diagnostic code related to CHF, registered by their primary care physician (I11.0, I13.0, I13.2, I50, I50.0, I50.1, I50.9, P29.0, according to the International Classification of Diseases Tenth Revision used in primary care) were included. As a measure of diagnostic accuracy we only included patients prescribed diuretics. In the urban region, we included the randomized set of PCPs that followed usual care plus clinical practice guideline on CHF and excluded those PCPs that underwent the disease management intervention. We included all PCPs in the rural region.

We measured survival after the onset of CHF until death or the end of follow-up (December 31st, 2007). Onset of CHF referred to the recorded date of diagnosis of CHF, which we extracted from the primary care electronic patient records. Registration of the diagnosis was done by primary care physicians and in our context is usually done after cardiologist consultation or advice (Table 1) or after the patient has been admitted to hospital. As a measure of the accuracy of the diagnosis, we only included patients prescribed diuretics. We also measured survival in patients with hospital admission for cardiovascular causes during the follow-up, as a primary discharge diagnosis (398-398.99, 402-402.91, 428-428.9, 997.1, 403.90-403.91, 404-404.93, 411-411.89, 414-414.9 according to the International Classification of Diseases Ninth Revision used by hospitals). We included mortality from any cause.

Potential comorbidities associated with survival were defined on the basis of the International Classification of Diseases Tenth Revision codes registered in the primary care electronic patient records either before or at the onset of CHF and were chosen according to a review of the literature, clinical relevance, and availability. We included patient age and sex, hypercholesterolemia, hypertension, diabetes mellitus, ischemic heart disease, chronic obstructive pulmonary disease, and chronic kidney disease (Table 2). We obtained age by calculating the difference between the date of inclusion in our study (January 1st 2005) and date of birth and considered 2 categories: patients aged less than 65 years and patients aged 65 years or older. Setting was also considered as a covariate.

From January 2005 to December 2007, we extracted data from 3 national database sources, which we were able to match because every Catalan citizen has a unique and anonymous identification number for health care use. The central database of the Catalan Health Service facilitated data abstraction from the electronic patient records of PCPs. Data from hospital admissions was gathered from the Division of Demand and Activity Registries (Minimum Basic Data Set for Acute-care Hospitals) of the Catalan Health Service, where all Catalan hospitals are required to send in their data for reimbursement. Information on mortality was gathered from the Mortality Register of Catalonia, where fatalities had to be notified. We recorded as fatality cases patients codified as such in the primary care electronic patient records register and those also identified in the Mortality Register of Catalonia.

None of the officers responsible for data abstraction were involved in the subsequent data analysis.

Descriptive data from our sample on age, sex and comorbidities are presented. We calculated the entire cohort follow-up period and reported the median, minimum and maximum values. Candidate variables significantly associated with mortality in the bivariate analysis (P<.05) were included as potential covariates in the multivariable model. In the bivariate model, the chi square text and Fisher's exact test were used for categorical variables and Student's t test for continuous variables. Kaplan Meier analysis was used to determine long-term survival. The Cox proportional hazards model was used to examine the independent effect of candidate variables. The forward and backward step technique with the likelihood ratio test was used. Discrimination of the model was calculated with Harrell's C-index, and calibration of the model was calculated with the Gronnesby and Borgan test. Constant hazard was tested with the log (−log S[x]) vs Log(time) graphic. All tests were 2-tailed and significance was set at the 5% level (α=0.05). Missing values were calculated (0.2% of our final sample) and were found not relevant in our analysis.

The statistical analysis was performed with SPSS v18 software and StataCorp12.

Our study has some potential design limitations concerning the inclusion of participants and measures. Although we did not have access to the set of PCPs excluded in the urban region, we found no significant differences among the regions included in this study. We expected that any other ethnic or socioeconomic effect on outcomes would have been minimized by our selection process, which developed from a previous randomization performed for a concomitant disease management intervention.

We identified CHF patients through their primary care physician's electronic patient records but did not know how many of them had undergone echocardiography and therefore lacked information on the etiology of CHF, which would have aided interpretation of our results. As a measure of diagnostic accuracy, we only selected patients who had been prescribed diuretics to perform our data analysis and possibly missed patients with CHF not receiving diuretics.

Usually, in our setting, primary care physicians register the diagnosis of CHF after confirmation by a specialist. This specialist is also involved in the diagnostic process and management of these patients in the community and provides support to primary care physicians in their setting, as this has been part of our integrated care program since 199034 (Table 1). Although we could not exclude the possibility that diagnosis of CHF was made clinically by the primary care physician alone in some cases, our results on survival were consistent with those found in the Spanish national data registry.

Our predictive model had a modest predictive power (c=0.67) and also low goodness-of-fit (P=.03), particularly for the first decile, with higher goodness-of-fit for the remaining deciles, indicating that the model failed in patients at lower risk. These results could be explained by 2 reasons: a) comorbidities were identified through the primary care physicians’ registration in electronic patient records and we were unable to capture how long patients had been exposed to the conditions or to obtain information on severity by using biomarkers, and b) the sample size was large (over 500 cases in each decile), increasing the possibility that every small difference would be significant. Nonetheless, our model was clinically relevant and in line with reports in the literature.

Because of the low registration on New York Heart Association classification and left ventricular ejection fraction, we could not present survival data according to these factors. Such information would have provided more detailed information on the severity of CHF. Primary care physicians do not usually record this information, nor do they do so as open access text, which also hampered access.

Finally, we considered the date of registration of CHF in the primary care physicians’ electronic medical records, which did not necessary reflect the date when the diagnosis was made, although our survival rates matched those in national database registers.