The DIOCLES is a multicenter, cross-sectional, quality controlled, observational study conducted in the first half of 2012, with the participation of 44 centers from 13 autonomous communities in Spain. The study prospectively collected data from patients aged ≥ 18 years consecutively admitted with suspected ACS. Demographic variables, risk factors, medical history, clinical presentation, complications, in-hospital mortality, and prehospital, in-hospital, and discharge management were recorded. At 6 months, the occurrence of death and its date and cause were determined by telephone interview. The study population comprised 2557 patients out of the 3059 evaluated, after 502 were excluded for various reasons.9

The 2012 MBDS contains 400 861 hospitalization episodes recorded in hospitals of the Spanish National Health System with a primary diagnosis of cardiovascular diseases or without that diagnosis and discharged from a cardiology or cardiac surgery department. The MBDS includes information on patients’ demographic characteristics and on variables related to the health care and clinical process, to patients’ diseases and conditions, and to the procedures performed during their care, coded by the International Classification of Diseases, Ninth Revision, Clinical Modification.15

The DIOCLES registry and the MBDS differ in their scope–DIOCLES includes 6-month follow-up after discharge, whereas the MBDS includes only the hospitalization episode–and in their data models. Given that these models lack common attributes that allow the records corresponding to the same episode to be unequivocally matched (direct identifiers), different indirect identifiers were tested to link them.32 The identifier found to obtain the most unique matches was selected. This model comprised the hospital code, dates of admission and discharge, and patient age and sex, because date of birth was not recorded in DIOCLES.

A descriptive analysis was performed of the variables studied; discrete variables are expressed as No. (%), and quantitative variables, whose normality was assessed by the Shapiro-Wilk test, as the mean ± standard deviation. To compare discrete variables, the chi-square test or Fisher exact test was applied, as necessary, with the Student t test or Mann-Whitney U test used for quantitative variables.

The Cohen kappa coefficient was used to evaluate interobserver agreement. Sensitivity, specificity, and positive and negative predictive values were calculated to measure the validity of the MBDS. Among the variables common to the 2 sources, this analysis included some of those most relevant for the adjustment of the risk of hospital mortality due to AMI12 (Table 1). To estimate the discrimination power of the MBDS, considering the values of the analyzed variables together, the area under the ROC (receiver operating characteristic) curve was used; the homogeneity of κ and of the area under the ROC curve were examined with the chi-square test. The areas under the ROC curve were compared by hospitals grouped into quintiles according to the number of patients included in DIOCLES. The dependence between observations in the same patient was analyzed by the intraclass correlation coefficient (ρ) and the variance inflation factor.34

The degree of agreement was interpreted according to the Landis and Koch scale.35 Statistical significance was defined by P < .05 and analyses were performed with Epidat v3.1 and STATA 13.0.

The impact of match failures on the results was evaluated through a sensitivity analysis by allowing in the linkage differences of up to 2 days in the admission and discharge dates between DIOCLES and the MBDS and by selecting, among multiple matches, the episodes showing fewer discrepancies among the conditions studied.

In addition, as an alternative linkage procedure, MBDS episodes with ACS as the primary diagnosis and an admission date in the first half of 2012 were selected, and the hospital code, age, sex, and dates of admission and discharge were used as indirect identifiers to match these episodes with those recorded in DIOCLES. On the results obtained through this procedure, a sensitivity analysis was also carried out following the same criteria as in the original procedure.

More than 85% of the diagnoses and procedures analyzed had a probability of being true positives if present in the MBDS and more than 98% had a probability of being true negatives if not. Thus, the present work indicates a threshold for the validity of the MBDS as a data source for the study of ACS in Spain, although its main limitation is the percentage of matches that could not be resolved.

Because the use of alternative linkage procedures did not reduce the percentage of failed matches, which ranged between 55% and 80%, the difficulty of identifying the episodes included in DIOCLES in the MBDS must be due to registration inaccuracies in either of the 2 sources. To our knowledge, the inclusion of age instead of date of birth in DIOCLES is a significant factor, but there may be others.

Dissociation is a legal requirement for personal data protection that prevents the use of direct identifiers to link CRs and administrative databases. To resolve this restriction, indirect procedures have been developed that achieved unique matches in proportions ranging from 58.1% (Setoguchi et al.33) to 87.5% (Pasquali et al.31), 87.9% (Austin et al.30), and 90.8% (Hammill et al.32). The reasons for the match failures32 have been described; although most are determined by the study context and cannot be applied to our situation, it is likely that, in DIOCLES, the date of patient arrival at or release from the emergency department or cardiology department was sometimes recorded as the date of patient admission or discharge, which might not coincide with that of the MBDS (which reflects when the patient is admitted to the hospital and occupies a hospital bed41). The reasons for the discrepancies found by Sarkies et al.42 between the use of administrative data and observational registries for the calculation of the average stay per hospitalization episode would support this explanation, consistent with the increase in the percentage of matches to practically 80% when registry differences of up to 2 days are allowed for the dates of admission and discharge.

In this case, the sensitivity analysis revealed that an increase in the number of cases by almost 32% did not significantly modify the concordance between DIOCLES and the MBDS, except for AMI and angioplasty. However, the concordance is still almost perfect, and there are only differences in the proportion of patients also with AMI and angioplasty in the MBDS. Thus, the risk of a selection bias as a result of mismatches seems unlikely and, even so, the sensitivity would still be around 80% and the specificity above 98%.

If the completeness of the sample is deemed important, the failure to detect all of the episodes recorded in the DIOCLES would affect the use of the MBDS in the study of ACS-related health care and could be due to match failures, an issue that should be resolved using the original clinical documentation.

The present work has other limitations. One limitation might be due to the participation in DIOCLES of a private hospital, whose episodes could not be included because they were not registered in the MBDS. Another might be due to the selection of the variables studied; however, because some of the main risk factors for the adjustment of in-hospital mortality due to AMI were considered as diagnoses and procedures, there seems to be no reason to believe that a different selection would have altered our observations. Another, more important, limitation is the result of not having used the primary information on the patients included in DIOCLES as reflected in the medical records, whose future comparison with the data sources used in this work is considered essential to verify the validity of the MBDS.

Irrespective of these considerations, the linkage of the CRs with the MBDS, as discussed here, points more to complementarity rather than to antagonism, to the extent that CRs can systematically benefit from the inclusion of previously validated administrative data (eg, patients’ date of birth instead of their age), setting a course that, due to the rapid growth in information technologies, will soon increase the availability of large repositories of clinical and administrative information for secondary uses related both to biomedical research and health care management.