Changing Mortality in Congenital Heart Disease

Data sources and collection

Quebec is the second most-populated province in Canada, with over 7.6 million inhabitants. A unique Medicare number is assigned to every individual, and all diagnoses and health services provided are systematically recorded over a lifespan (1). The physician's services and claims database contains demographic data and records on all patients who contacted the health care system, as inpatients or outpatients, and includes diagnostic and therapeutic procedures performed since 1982. Diagnostic codes adhere to the International Classification of Disease-Ninth Revision (ICD-9). As physician remuneration is based on these claims, there is a strong incentive to divulge all professional activities. The hospital discharge summary database contains records on all hospital admissions and captures demographic data, diagnostic codes, and therapeutic procedures. All diagnoses are linked with specialty-specific provider codes.

Data on all patients with a diagnosis of CHD between January 1, 1983, and June 30, 2005, who were alive on or born after July 1, 1987, were obtained from the Quebec Health Insurance Board. To protect patient confidentiality, information was transmitted with encrypted identifiers. Using the same encrypted numbers, the Quebec Ministry of Health provided additional hospitalization discharge data. These source databases were linked for each individual, thereby generating a comprehensive record of health care coverage for patients with CHD. The study was approved by the McGill University Health Center ethics board and the Quebec government agency responsible for privacy of access to information.

Study population

A hierarchical categorization scheme, summarized in Table 1,was used to classify patients with CHD (1). We identified all patients with at least 1 diagnostic code for CHD and/or surgical procedure specific to these codes who came into contact with the health care system. The 24 ICD-9 diagnostic codes and corresponding surgical codes were grouped into 5 categories of decreasing specificity (Table 1) and cross-referenced between source administrative databases.

Using provider codes, physicians reporting diagnostic codes or procedures were classified as cardiovascular specialist (pediatric, medical, or surgical), primary care physician (general or family physician, pediatrician, or internist), echocardiographer, or other (1). When more than 1 ICD-9 code was recorded, the most frequent diagnosis in the diagnostic hierarchical block in Table 1with the lowest numerical value was retained, prioritizing diagnoses by cardiovascular specialists. We adhered to the guiding principle of minimizing, to the greatest extent possible, subjective opinion in categorizing lesions. Thus, each diagnostic category encompasses a spectrum of severity, with no judgment as to physiologic or clinical ramifications. Lesions included in block 1 of Table 1were considered “severe,” because these diagnoses were most likely to be associated with cyanosis or require surgical intervention in early life (1). To minimize misclassification errors, this systematic algorithmic approach was subject to sensitivity analyses of randomly selected samples within each diagnostic category (1,8). All discrepancies were reviewed and adjudicated by 2 independent CHD specialists.

Study design

A retrospective open-cohort study was conducted, with a maximum follow-up of 18 years, by identifying all patients with CHD alive each year from July 1, 1987, to June 30, 2005. Given that congenital disease is present at birth, patients with any diagnostic code for CHD between January 1, 1983, and June 30, 2005, entered the study cohort on July 1, 1987, if alive and born prior to that date, or at birth if born after that date.

Mortality data

In Quebec, death certificates must contain nominal and demographic data, along with the individual's unique and unvarying Medicare number. By law, the attestation of death must be sent to the registrar of civil status who then forwards this information to the Quebec Health Insurance Board. Further validation of mortality data occurs by cross-referencing vital status information provided by other Quebec institutions. Documentation of death is theoretically complete in this database, whether it occurs in or out of hospital. In addition, the hospital discharge database was used to cross-reference in-hospital deaths. Since this database was initiated in 1987, the beginning of follow-up was defined as July 1, 1987, to optimize data accuracy and allow fiscal years to be compared with general Quebec population data. With this strategy, originally developed to maximize reliability of vital status information for patients without CHD, death could be confirmed in 99.75% of cases (9).

Statistical analysis

In descriptive analyses, the distribution of age at death among subjects with CHD who died in a given fiscal year, starting from July 1, 1987, was characterized by histogram plots and medians and interquartile ranges (IQRs) (25th, 75th percentiles). Median ages at death in selected years were compared with Kruskal-Wallis tests.

Statistical analyses focused on changes in mortality rates across six 3-year time intervals, from 1987 to 1990, to 2002 to 2005. These analyses accounted for the fact that some subjects were identified only at time of death and, thus, similar subjects who did not die in the same period were not identified, or were identified only much later in the database. To avoid resulting biases in estimating mortality rates, statistical analyses were, therefore, limited to restricted cohorts. Restricted cohorts were defined as the population of subjects indentified prior to the beginning of each 3-year interval who remained alive (i.e., “at risk”) at the beginning of the interval. For every subject in a given restricted cohort, death within the corresponding 3-year period was modeled as a binary dependent variable. Adjusted mortality rates in the periods 1990 to 1993, to 2002 to 2005 were compared with the earliest period (1987 to 1990) with multivariate Poisson regression models that controlled for age at the beginning of the period, sex, and subtype of CHD categorized as severe, simple shunt, and other. Since individual patients may have been included in more than 1 restricted cohort, analyses were performed with and without the generalized estimating equations (GEE) extension of Poisson regression (10) to account for the dependence of within-subject observations (11). Since the 2 approaches yielded similar results, we report adjusted mortality rate ratios and their 95% confidence intervals (CIs) without GEE. Poisson regression analyses of all CHD lesions were performed separately for 6 age groups. For each age group, adjusted mortality rate ratios were likewise estimated from general Quebec population data using Poisson regression adjusted for sex and age within each group (except infants). In lesion-specific analyses, to ensure sufficient statistical power, patients were regrouped as children (age <18 years) or adults (age 18 to 65 years), and less frequent CHD subtypes were not considered.

Two-tailed p values <0.05 were considered statistically significant. All analyses were performed using SAS software version 9.1 (SAS Institute, Cary, North Carolina).

Study population

The full cohort was comprised of 71,686 patients with CHD followed for 982,363 patient-years. The general Quebec population varied between 6.8 and 7.5 million people during the first (1987 to 1988) and final (2004 to 2005) years of follow-up, respectively. Characteristics of the full and restricted cohorts are summarized in Table 2.

Distribution of age at death in CHD

A total of 8,561 (11.9%) deaths occurred during follow-up. As portrayed by the histogram on the left of Figure 1,the distribution of age at death during the first year of observation (1987 to 1988) displayed a bimodal pattern, with the highest mortality peak in infancy and a lower peak in later adulthood. In contrast, by the last year (2004 to 2005), the distribution shifted toward deaths at older ages, becoming nearly unimodal (right-hand graph in Fig. 1). In 1987 to 1988, 49.0% of deaths in patients with CHD occurred under the age of 20, whereas by 2004 to 2005, only 9.3% of deaths occurred under age 20 years. Although in 2004 to 2005, deaths at ages 0 to 4 years remained somewhat more frequent than in later childhood, the distribution of age at death after 4 years of age approximated that of the general Quebec population (bold curve).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Distribution of Age at Death in Patients With Congenital Heart Disease in 1987 to 1988 and 2004 to 2005

Histogram bars depict the proportion of all deaths (x-axis) according to age at death (y-axis) in our cohort of patients with congenital heart disease in the first (1987 to 1988; left) and final (2004 to 2005; right) years of observation. Bold black curves with diamondsrepresent the corresponding age at death distribution in the general Quebec population during the same periods of observation.

Temporal changes in mortality

Overall, the median age at death increased by 15 years, from 60 years (IQR: 3 to 76 years) in 1987 to 1993, to 75 years (IQR: 60 to 83 years) in 1999 to 2005 (p < 0.001). In patients with severe forms of CHD, the median age at death increased from 2 years (IQR: 1 to 13 years) in 1987 to 1993, to 23 years (IQR: 1 to 56 years) in 1999 to 2005 (p < 0.001). Table 3summarizes the number of deaths in each restricted cohort and the proportion of deaths that occurred in patients with severe versus other forms of CHD, and in children versus adults.

Figure 2shows how adjusted mortality rates varied over time for each age category, relative to the first 3 years of observation. The mortality reduction in children with CHD was greatest in infants, became smaller with increasing age, and exceeded that of the general population. Infants with CHD experienced a 77% reduction in mortality (mortality rate ratio: 0.23, 95% CI: 0.12 to 0.47, p < 0.001), in comparison to a 52% decline (mortality rate ratio: 0.48, 95% CI: 0.38 to 0.60, p < 0.001) in the general infant population. For children 5 to 17 years of age, the mortality reduction was 66% (mortality rate ratio: 0.34, 95% CI: 0.21 to 0.54, p < 0.001) in the CHD cohort compared with 50% (mortality rate ratio: 0.50, 95% CI: 0.45 to 0.55, p < 0.001) in the general population.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Adjusted Age-Specific Changes in Mortality Over Time in Children and Adults With Congenital Heart Disease

For each of the 6 age categories (6 panels), adjusted mortality rate ratios (black squares)and 95% confidence intervals (vertical lines)are plotted relative to the reference cohort (1987 to 1990). Mortality rates are based on restricted cohorts (see Statistical Analysis section) and were estimated by Poisson regression models that controlled for residual age differences, sex, and type of congenital heart disease. Sex was not available for infant deaths that were captured by a temporary administrative database prior to allocation of the permanent unique identifier. The difference in mortality is statistically significant at a 2-tailed alpha of 0.05, if the corresponding vertical barfor the 95% confidence interval does not cross the horizontal dotted line(corresponding to a mortality rate ratio of 1).

In all but the oldest age group, adults with CHD experienced statistically significant reductions in mortality that paralleled general population trends. Mortality rate ratios for adults 18 to 34 years of age with CHD and in the general population were 0.60 (95% CI: 0.40 to 0.90, p = 0.02) and 0.64 (95% CI: 0.61 to 0.67, p < 0.001), respectively. For adults 35 to 64 years of age, corresponding mortality rate ratios were 0.72 (95% CI: 0.55 to 0.93, p = 0.02) and 0.70 (95% CI: 0.69 to 0.71, p < 0.001). Mortality rates declined in the general population of adults age ≥65 years (mortality rate ratio: 0.78, 95% CI: 0.77 to 0.78, p < 0.001), but not in the CHD cohort (Fig. 2).

Table 4provides adjusted mortality rate ratios, according to CHD subtype, separately for all patients, children, and adults 18 to 64 years old. Gains in survival were mostly driven by decreased mortality in severe forms of CHD, particularly in children who experienced a 67% reduction in mortality (mortality rate ratio: 0.33, 95% CI: 0.19 to 0.60, p < 0.001).

Study limitations

The targeted catchment population was the entire province of Quebec. However, subjects with CHD who had no contact with the health care system from January 1983 to June 2005 were not captured. This is more likely to affect those with mild forms of disease. Additionally, the open cohort design carries a higher risk for missing subjects with a shorter time frame to establish a diagnosis of CHD, either because they were born closer to the end of the study period or died shortly after the cohort inception. However, this potential bias should not have a sizeable impact on observed changes in ages at death since median values are quite robust with respect to the potential omission of a relatively small fraction of the total population. To avoid potential biases, our primary analyses of mortality rates were limited to restricted cohorts of subjects “at risk” at the beginning of each observation period (i.e., employed the same criteria to identify the population in the “denominator” and deaths in the “numerator”). Furthermore, since selection bias may result from a higher likelihood of detecting milder forms of CHD in more recent calendar years, we adjusted for CHD subtype. In addition, the proportion of patients who died, but were not included in restricted cohorts, were compared in an exploratory analysis. Since the proportion of excluded deaths stayed relatively constant over time, one could reasonably conclude that the selection effect had not changed substantially over time. To further address the potential for left-censoring bias, we conducted an analysis that restricted inclusion to patients diagnosed with CHD within 4 years of the start of each successive cohort. Despite the exclusion of 42% of deaths, this sensitivity analysis yielded congruent results (e.g., overall adjusted mortality rate ratio 0.69 [95% CI: 0.62 to 0.77] in comparison to 0.69 [95% CI: 0.61, 0.79]).

Our diagnostic classification system was limited by the ICD-9 codes. This coding system was previously used to quantify prevalence rates, births, and deaths in CHD (1,13,40). Given our reliance on administrative databases, diagnostic misclassification errors remain possible despite every effort to maximize data accuracy. However, misclassification errors are likely nondifferential such that any resulting potential bias would be towards the null (41) and could not, therefore, explain the observed statistically significant decreases in mortality. Finally, the impact of socioeconomic status was not assessed in the current analysis, and ethnic origin is not available in our concatenation of data resources. Nevertheless, racial variation, as reflected by the proportion of visible minorities in census bureau statistics, is substantially less in Quebec compared with the U.S. population.