Methods

Study design and inclusion criteria

The Experimental Prospective Study on the Effectiveness and Efficiency of the Implementation of Clinical Pathways was designed as a cluster randomised controlled trial, in which the experimental group was composed of HF patients treated according to a specific CP, while the control group received usual care. The methodology of the study has been described in detail previously.9 In brief, we used a cluster design (with single hospitals as randomisation units) because of the ethical and logistical issues associated with the implementation of CP, which involves a series of complex actions at the institutional level.101112 Forty hospitals located in four Italian regions were invited to participate; 18 of them showed interest in the implementation of CP for HF and were assessed for eligibility. Based upon the comparability of hospital location, patient population, facilities and teaching status, 14 hospitals were selected and randomised. All data were prospectively collected from March 2003 to October 2004 by physicians and nurses who were trained in prestudy educational events. The sample included all the patients treated by the hospitals with a principal diagnosis of HF (all International Classification of Diseases, Ninth Revision, Clinical Modification codes included in 428.xx code) and patients with acute myocardial infarction or unstable angina were excluded.13 The project was approved by the http://ClinicalTrials.gov and was exempted from ethical clearance according to the Italian Ministry of Health law (ex art. 12bis D.lgs 229/99).

Outcome measures

The primary outcome measure was in-hospital mortality. Secondary outcomes were length and appropriateness of the stay (measured using the Italian version of the Appropriateness Evaluation Protocol),14 unscheduled readmissions rate (based upon Italian administrative database, SDO); customers’ satisfaction and costs of hospital care. The quality of care was assessed by monitoring the use of diagnostic and therapeutic procedures during hospital stay and by the use of quality indicators at discharge.15161718 The indicators are reported in table 1.

Intervention

One physician or nurse with at least 2 years of experience of CP was assigned to each hospital in the experimental group, in order to facilitate project implementation (including staff education in the use of CP). The teams included internal medicine physicians, cardiologists, epidemiologists, pathologists, psychologists, nurses, hospital pharmacists, social workers and support staff. The teams were formed on a voluntary basis, received three days of training in the development of CP and constructed the CP over a 6-month period. All groups analysed their care processes, reviewed best evidence provided by senior investigators, defined the appropriate goals of the pathways, detailing the results into protocols and documentation, including the sequence of events and expected progress of the patients over time.615 Essentially, the CP used in the study were not completely identical because of organisational adaptations in some sites. However, they coincided substantially with the existing European guidelines on the hospital treatment for HF.19

Data collection and statistical analysis

Sample size calculation was performed according to standard criteria for cluster randomised trials.122021 In brief, the intracluster correlation coefficient for in-hospital mortality was estimated equal to 0.018, leading to a design effect of 1.51. Using standard 0.05 α error and assuming a 15% difference in mortality between the intervention and control group,2223 and 10% of losses to follow-up, a sample size of 424 patients (212 in each group) was required to obtain a statistical power of 0.80. The seven clusters in each arm were thus comprised of 31 patients.

The Fisher exact and Kruskal-Wallis tests for categorical and continuous variables, respectively, were performed at the cluster level. In addition, differences in the rate of in-hospital deaths and unscheduled admissions between groups and according to each variable under study were evaluated at individual level using random-effects logistic regression, accounting for the clustering effect.2425 Variables were included if significant at the 0.10 level (backwards approach), with the exception of age which was forced to entry. The presence of multicollinearity, interaction and higher power terms was assessed to check final model validity. Patients who died during the study were excluded from the regression model predicting unscheduled readmissions because they could not be readmitted. Statistical significance was defined as a two-sided p value <0.05. All analyses were intention-to-treat and were carried out using STATA statistical software, version 8.2 StataCorp LP 4905 Lakeway Drive College Station, Texas 77845 USA.

Results

The final sample consisted of 429 patients (214 receiving CP, 215 usual care). At admission, there were no significant differences between the groups according to gender and selected clinical conditions, while the mean age was slightly higher in the experimental group (table 2). Although only patients with New York Heart Association (NYHA) III–IV scores or higher had to be admitted, a minority of patients with NYHA II scores were also admitted because of low socioeconomic status (7%).

The discharge status of all patients, according to the type of treatment received, and several outcomes of care are reported in table 3. In-hospital mortality was 5.6% in the group receiving CP; 15.4% in the control group (p = 0.001). At bivariate analysis, all secondary outcomes showed better results in the experimental group (proportion of appropriate days of stay; costs of the care and patient’s satisfaction), although only the length of stay and the rate of unscheduled readmissions approached significance.

The results of the multivariable random-effect logistic models predicting in-hospital mortality and unscheduled readmissions are shown in table 4. Adjusting for age, smoking, NYHA score, hypertension and source of referral (general practitioner or other), patients who were managed and cared for according to CP, as compared to subjects in usual therapy, had a significantly lower risk of in-hospital death (OR 0.18; 95% CI 0.07 to 0.46) and unscheduled readmissions (OR 0.42; 95% CI 0.20 to 0.87).

With regard to the use of diagnostic procedures, echocardiography, oximetry and diuresis monitoring were much more frequently utilised for patients treated according to the CP as compared to controls (p<0.005, table 5). Similarly, all medications were administered more frequently in the experimental group, with the exception of diuretics and anti-platelet agents.

Mean values, for each group, of the four quality indicators recorded in the study are shown in table 6. Compared with controls, the proportion of patients receiving left ventricular function assessment, advice/counseling on smoking cessation and written discharge instructions was significantly higher in the experimental group, while ACE inhibitors at discharge were more frequently requested in the CP group, however this difference was not significant.

Discussion

The main finding of this study is that care delivered using CP based on current guidelines, as compared to usual care, is more effective in reducing in-hospital mortality in HF patients. Furthermore, the implementation of such a clinical governance tool resulted in a significant improvement of most outcomes of care, including the rate of unscheduled readmissions, length of stay, diagnostic procedures and medication use, and most JCAHO quality indicators.

There may be a number of explanations for the observed reduction in in-hospital mortality, which are not mutually exclusive and which include (a) improved performance, (b) earlier discharge of palliative patients and (c) a baseline difference between the two groups in terms of vital signs, laboratory data, and/or co-morbidities, due to selection bias. Concerning the latter, after randomisation the two groups were similar for co-morbidities and NYHA scores. In addition, CP use remained a strongly significant determinant of mortality after adjusting for several potential confounders. Therefore, although we could not use the ADHERE risk stratification model,252627 because the study was initiated before its publication, the influence of selection bias, if any, is likely to be minor.

The second possible explanation for the observed findings is a different attitude between the groups towards the care of patients close to death. Indeed, if the hospitals in the experimental groups discharged patients when close to death, these would have artificially reduced their in-hospital mortality rate. However, although we did not have reliable data on the events after discharge, the average length of stay in CP was only 1 day lower than in usual care and, more importantly, the severity at discharge according to NYHA classes was lower in the experimental group.

In light of the above considerations and in agreement with other authors,15 it is thus reasonable to believe that the observed reduction in in-hospital mortality should be attributed to an improvement of the quality of care determined by the use of CP.

With regard to secondary outcomes, our results were consistent with the findings of the OPTIMIZE-HF programme, which demonstrated how evidence-based care can be effectively implemented in real-world patients hospitalised with HF.28 Indeed, the four JCAHO performances measures were improved in the CP group,5 in which we detected also a significantly better diagnostic and therapeutic pattern.234 These findings were observed despite the complexity of an objective evaluation of the processes of care because of dependence on clinical practice, susceptibility to inadvertent bias and lack of validated measures.

No differences were found in admission organisational appropriateness and in patients’ satisfaction scores, which, however, were high in both groups at discharge. The most likely explanations include the use of short-term outcomes and an insufficient statistical power for these indicators. An alternative approach based on the evaluation of quality of life at 6 months might have shown different results.29

Given the type of intervention, a cluster randomised design was probably the most appropriate design to use.30 However, these studies have some limitations when applied to CP.31 Indeed, the context level adaptation, which is essential for the pathways to succeed, may be perceived as inappropriate in the trial design and it may be difficult to replicate and maintain the original intervention. To reduce such issues we implemented pathways that combined local standards with evidence-based indicators, in order to maintain the integrity of the intervention in each site.32

Another potential limitation of the study was related to the information system, especially in the method of documenting and collecting data from current sources (clinical records, paper-based abstraction tools). Certainly, the use of a national registry like ADHERE and OPTIMIZE-HF may provide, in the future, a better evaluation of methodologies such as CP. Moreover, the cost of development and implementation of CP has not been evaluated, thus we cannot conclude that the implementation of our care pathway was a cost-effective process.

Finally, a recent study criticised the relationship between current HF performance measures and patient mortality or combined mortality/rehospitalisation in the first 60–90 days after discharge.33 However, other studies supported the use of the same measures,23435 and within this study short-term outcomes such as in-hospital mortality and readmissions within 31 days were utilised. Furthermore, after risk adjustment CP remained strongly significant determinants of reduced in-hospital mortality and unscheduled readmissions.

In conclusion, the present study adds to the evidence suggesting that the implementation of CP for the hospital treatment of HF produces a significant improvement in the quality of care and reduces in-hospital mortality. Further efforts should be devoted to enhance the diffusion of CP based upon reliable evidence.