Temporal Trends in Infective Endocarditis:  A Population-Based Study in Olmsted County, Minnesota FREE

The representativeness heuristic of the clinical features of infective endocarditis (IE) as described by Osler has undergone a significant change in developed countries.^1- 2 Previously, IE was a disease that commonly affected patients with predisposing valvular abnormalities caused by rheumatic carditis, with viridans group streptococci the most common causative pathogens.³ This presentation is currently seen in developing countries, where rheumatic heart disease is still prevalent. In developed countries, mitral valve prolapse (MVP) is now thought to be the most common predisposing cardiac condition in patients with IE.² Several recent studies from passively reported case series suggest that Staphylococcus aureus is now the most frequently identified causative pathogen.¹ These more recent clinical observations of IE characteristics, however, are based on data that come primarily from large, tertiary care centers^4- 5 and may not reflect true changes in the epidemiology of IE but rather temporal changes in referral patterns. Thus, population-based investigations are needed to more accurately characterize IE in the United States.

Olmsted County, Minnesota, is a setting uniquely qualified for the conduct of population-based studies of disease epidemiology. The Rochester Epidemiology Project has been continuously supported by the US federal government for more than 30 years⁶ and has played a pivotal role in the examination of disease epidemiology. Key features of the project and the county include⁶ that: (1) the local population is relatively isolated from other urban centers and receives the majority of its care from a small number of health care practitioners; (2) medical care is largely self-contained within the community; (3) medical, surgical, and pathological diagnoses are identified in a computerized central index covering essentially all sources of medical care (both inside and outside of Mayo Clinic) used by county residents; and (4) the original medical records of all inpatient and outpatient care are readily available for review. Thus, essentially complete ascertainment of IE cases in Olmsted County is possible and all details of medical care provided to the residents are available for study.³ We therefore conducted a community surveillance study of patients with IE from Olmsted County during the period 1970-2000. We evaluated trends in the incidence and clinical characteristics of IE that occurred in a population-based cohort spanning 31 years.

The characteristics of Olmsted County are similar to those of US non-Hispanic whites.⁶ This population consists largely of middle-class whites, with a low prevalence of injection drug abuse.⁷ The Mayo Clinic and Olmsted Medical Center provide the majority of medical care for the population as developed in geographic isolation from other urban centers. Today, the closest competing medical centers are in Minneapolis, Minn (139.2 km to the north), LaCrosse, Wis (113.6 km to the east), Iowa City and Des Moines, Iowa (316.8 and 332.8 km to the south, respectively), and Sioux Falls, SD (376 km to the west). Although best known as a tertiary referral center, the Mayo Clinic has always also provided primary and secondary care to local residents. Because Mayo Clinic and Olmsted Medical Center offer care in every medical and surgical specialty and subspecialty, local residents are not obliged to seek clinicians throughout a large region but are able to obtain most of their health care within the community.⁶

Because virtually all clinicians in Olmsted County participate in the Rochester Epidemiology Project, any endocarditis case that was treated in Olmsted County would have almost certainly come to our attention. Even for Olmsted County residents who are diagnosed and treated elsewhere, upon return to the county the diagnosis will be noted in their medical records unless the patient dies elsewhere. Furthermore, the population is relatively stable, particularly among the older age groups. For example, the median length of follow-up available for residents 50 to 59 years of age is 29 years. For these reasons and because of the rarity and severity of IE, it would be very unusual, albeit possible, for an IE case to be missed.

Olmsted County medical institutions use a unit medical record system in which information is collected by health care clinicians in a single record for each patient. These records are easily retrievable through the Rochester Epidemiology Project, which links and indexes diagnostic and procedure information from virtually all sources of health care in the county into a single centralized system. Adult (≥18 years) cases of IE among Olmsted County residents were identified using this system. Cases that were diagnosed at autopsy were also included. In addition, we used prospectively collected records maintained by the Division of Infectious Diseases at Mayo Clinic of all patients with IE seen in consultation by division members. The medical records of all possible cases were then reviewed to confirm the diagnosis of IE. All cases judged problematic were reviewed with a more experienced investigator (J.M.S.).

New IE episodes that occurred during the study period in patients with a previous history of endocarditis were included; 3 cases of posttreatment IE relapse were excluded. All potential cases were screened according to specific case definition listed below, and 107 episodes of endocarditis were identified for study. The institutional review boards at Mayo Clinic and Olmsted Medical Center approved the study and waived the requirement for informed consent.

Case definitions of IE have changed over time and include both the widely used Beth Israel⁸ and Duke⁹ criteria. Cases were identified in the current investigation by the modified Duke criteria.¹⁰ Several approaches were taken to ensure complete ascertainment of all IE cases. First, we screened all IE cases that were defined before the introduction of the Duke criteria in 1994. In addition, we sought cases that were rejected because they did not fulfill the Beth Israel criteria but were considered likely IE cases and administered empirical therapy for IE.

One of the investigators (I.M.T.) reviewed all cases and classified them according to the modified Duke criteria, the validity of which has been previously demonstrated.¹⁰ To assess the reliability of our case-classification procedure, another senior investigator (L.M.B.) independently reviewed and classified 20 IE cases selected randomly. There was 100% agreement on case classification.

We performed several steps to ensure reliability and inclusion of good-quality data. First, we used a standardized data abstraction form with detailed definitions of the variables. Demographic data as well as clinical, laboratory, and outcome data were abstracted from the complete (inpatient and outpatient) medical records using a precoded data collection instrument. Variables were obtained from thorough review of the medical record, including daily physicians’ progress notes and all subspecialty consultations. Nearly all patients were followed up daily in the hospital by infectious diseases and cardiology subspecialists, with additional consultations when indicated. Any uncertainties in data abstraction were discussed with an experienced investigator (J.M.S. or L.M.B.).

Ascertainment of death at 6 months was obtained from medical records in the majority of cases and was supplemented by the recording of all obituaries and notices of deaths by the Mayo Clinic registration office and the National Death Index.

For calculating incidence rates, the adult population (≥18 years) of Olmsted County was considered to be at risk; the denominator age- and sex-specific person-years were derived from decennial census figures. Intercensus figures were estimated using high-degree polynomial interpolation. Ninety-five percent confidence intervals (CIs) for IE rates were estimated assuming that the incidence cases follow a Poisson distribution. Rates were age- and sex-adjusted to the population of whites in the United States in 2000. A Poisson regression model (GENMOD procedure, SAS version 8; SAS Institute Inc, Cary, NC) was used to examine the temporal trends in the incidence of IE, with categorical year variables. Exact Wilcoxon tests for the ordered contingency table, in the case of categorical characteristics, were used to study time trends. The Spearman rank correlation coefficient was used to assess whether there was a significant temporal trend in age at diagnosis.

Definite and possible IE cases as defined by the modified Duke criteria were used in the analysis. A sensitivity analysis excluded all possible cases and included only definite cases in order to examine the effect of disease misclassification. We also performed a second sensitivity analysis that included only native valve IE cases. All analyses were conducted in SAS version 8. The level of significance for all statistical tests was 2-sided, with P<.05.

One hundred seven IE episodes occurred in 102 patients during the study period. The characteristics of the cohort are summarized in Table 1.

Table 2 summarizes the distribution of underlying heart disease, risk factors, and causative microorganisms. Mitral valve prolapse was the most frequent underlying valvular heart disease. Eighty-six percent of IE cases were caused by either streptococcal or staphylococcal species. Viridans group streptococci were the most common causative organisms and were isolated in 47 cases (44%); S aureus was identified in 28 cases (26%).

The Olmsted County adult population ranged from approximately 51 000 in 1970 to 90 000 in 2000. The overall average crude IE incidence for the period 1970-2000 was 4.95 per 100 000 person-years. The age- and sex-adjusted annual incidence was 6.06 per 100 000 (95% CI, 4.89-7.22).

Incidence. The overall adjusted incidence of IE ranged from 5.0 to 7.0 cases per 100 000 person-years during the study period and did not change significantly over time (P = .42 for trend) (Figure 1). Infective endocarditis continues to be a disease of older individuals, with a mean age ranging from 54.1 years in 1980-1984 to 67.4 years in 1995-2000 (P = .24 for trend). There was a male predominance (67%-83%), which did not significantly change over time (Table 3).

Causative Organisms. Viridans group streptococcal IE was the most common organism-specific IE subgroup, with an annual adjusted incidence of 1.7 to 3.5 cases per 100 000 person-years. Staphylococcus aureus IE had an annual adjusted incidence of 1.0 to 2.2 cases per 100 000. No time trend was detected for either pathogen group (P = .63 and P = .66, respectively) (Figure 2).

Underlying Heart Disease and Infected Valves.Table 3 summarizes the time trend of different IE characteristics. There was a nonsignificant declining trend over time in the proportion of cases with underlying rheumatic heart disease (P = .08). The proportions of cases with MVP and congenital heart disease did not change significantly over time (P = .19 and P = .82, respectively, for trend). In the subgroup of IE cases with identified underlying heart disease, there was a significant increasing trend in MVP over time (P = .04) and a decreasing trend in rheumatic heart disease (P = .08). However, the absolute numbers were small. The proportions of IE cases with aortic valve involvement (19%-54%) decreased over time (P = .02). An increasing trend was detected in the proportions of IE cases with prosthetic valve involvement (P = .09).

Outcome. There was no significant time trend in rates of valve surgery or in 6-month mortality over the 31-year study period (P = .97 and P = .59, respectively). Six-month mortality rates ranged from 14% to 33% (Table 3).

Limiting the analysis to Duke criteria definite IE cases did not substantially change incidence compared with when both definite and possible cases were included. The adjusted IE incidence ranged from 3.2 to 6.6 cases per 100 000 person-years, with the lowest incidence in the period 1970-1974. During the period 1970-1974 and before the introduction of echocardiography, one half of the cases (n = 6) were excluded because they were designated Duke criteria possible (Table 4). Viridans group streptococci remained the most common cause of IE, followed by S aureus. There were no trends over time in IE incidence caused by either organism. Similar findings were observed when the analysis was restricted to native valve IE cases (Table 4).

We postulated that transesophageal echocardiography, which was introduced in the mid 1980s, might affect any observed trend over time because of detection bias. Therefore, we compared the different incidences between the 1970-1984 and 1985-2000 periods. There was no significant difference in overall or organism-specific IE incidence (Table 5).

There were 4 IE cases in 2001 (2 due to viridans group streptococci, 1 due to Cardiobacterium hominis from the HACEK group, and 1 culture-negative), with an annual adjusted incidence of 4.65 (95% CI, 2.34-9.23). A complete enumeration of all IE cases in the county after 2001 was not possible because capture of all diagnoses through the medical records linkage system was not completed at the time of study.

The Rochester Epidemiology Project provides a unique opportunity to investigate trends in IE epidemiology among residents of Olmsted County, Minnesota. In this geographically defined community, the incidence of IE has remained stable during the past 3 decades. The adjusted incidence of IE ranged from 5.0 to 7.0 cases per 100 000 person-years. The broader diagnostic criteria that include Duke definite and possible classifications identified slightly more IE cases than did the stricter Duke definite classification alone, but the incidence remained stable, irrespective of the diagnostic criteria used. The adjusted annual incidence was 5.4 per 100 000 (95% CI, 3.7-7.2) for the period 1970-1984 and 6.5 per 100 000 (95% CI, 5.0-8.1) for the period 1985-2000 (P = .29). A recent study from France with a larger sample size observed a small decrease in the annual incidence of definite IE, from 3.1 cases per 100 000 in 1991 to 2.6 per 100 000 in 1999.¹¹ However, it was not designed to study temporal trends in IE.

Our results support the recent conclusion that despite improvements in health care over the past 30 years, the incidence of IE has not decreased.¹ At least 2 potential factors may contribute to this finding. First, there may have been a true overall decline in IE incidence, which was concealed by a detection bias with better blood culture techniques and more frequent use of echocardiography. Second, an increase in incidence caused by a more frequent use of echocardiography may have been offset, in part, by a declining number of autopsy-diagnosed cases.¹²

We observed that IE continues to be a disease predominately affecting older men, a finding consistent with other studies.^11,13- 14 In the last 5 years of our study, the majority of IE cases occurred in patients with no known underlying heart disease. This finding is consistent with a recent study.¹¹ There was a declining trend in proportion of rheumatic heart disease and an increasing trend in prosthetic valve and MVP proportions over time, although these trends did not reach statistical significance. However, absolute numbers were small, which made the proportions unstable and could have led to a spurious observation of apparent temporal trends as well as inadequate power to detect true ones.

We also observed that the incidence rates of viridans group streptococcal and S aureus IE have not changed significantly over time. Other groups have described an increasing frequency of S aureus IE^15- 16 or a decrease in viridans group streptococcal IE,^11,14,17 leading to a general consensus that S aureus has surpassed viridans group streptococci as the leading cause of IE.¹ In contrast, we found that viridans group streptococci continue to be the most common cause of IE in the study population and that its incidence rate is approximately twice that of S aureus. There are several possible explanations for these discrepant findings. First, differences in underlying at-risk populations may modify the organism-specific incidence rates. For example, communities with a high prevalence of intravenous drug use may observe an increased incidence of S aureus endocarditis. The rate of intravenous drug use is low in our population compared with that in other US populations, in which the rate of intravenous drug use among some IE series is as high as 18%.^15,18

A second possible explanation is underdiagnosis of IE among Olmsted County patients with S aureus bacteremia. A high proportion of S aureus IE cases can be missed clinically and are only diagnosed at autopsy or with routine use of transesophageal echocardiography in all patients with S aureus bacteremia. In a national study of all S aureus IE cases in Denmark between 1982 and 1991, IE was not suspected clinically and was only diagnosed at postmortem evaluation in 32% of the cases.¹⁹ Nevertheless, because medical care is largely self-contained within the Olmsted County community, patients with undiagnosed S aureus IE would ultimately seek medical care at one of the Olmsted County hospitals.

Third, and most likely, differences may arise due to different study design. The apparent increase in the proportion of IE cases due to S aureus in passively reported series from referral centers may be artifactual due to changes in medical practice that affect likelihood and patterns of referral.⁷

Our study has several important strengths. To our knowledge, it is the first population-based study that examined trend changes of IE characteristics in the US population over the past 2 decades. We used several approaches to overcome different biases, including case ascertainment bias and disease misclassification, that may have impacted the calculation of IE incidence in other work.

The essentially complete ascertainment of all IE cases in our study by active search in a well-established medical record linkage system for a population of known size and age distribution allows an unbiased and accurate estimation of the IE incidence rate. Case ascertainment is a major problem in studies of IE. Reported population-based studies have used 3 major types of study design to identify all IE cases and include survey design^{11,15,18,20- 21} and national discharge registries,^13,22 which are affected by the voluntary nature of notification or a lack of standardized diagnostic criteria and therefore incomplete case ascertainment.

We used 2 different sets of screening criteria to overcome disease misclassification, which is another type of bias in population-based studies of IE. Disease misclassification can be related either to case definition or to level of diagnostic certainty. First, we applied a unique case definition (ie, the modified Duke criteria) retrospectively to all identified cases to study trends in IE. Second, a sensitivity analysis was conducted that excluded all possible cases and included only definite cases to examine the effect of disease misclassification.

There are potential limitations to our study. First, the diagnosis of underlying heart disease was obtained from medical records and was based primarily on echocardiography and physician assessment without consistent pathological confirmation. However, pathological ascertainment is impossible in all cases because not all patients undergo valve surgery, die, or undergo a postmortem evaluation. Second, the relative uniformity of the racial and ethnic composition of Olmsted County and the low prevalence of intravenous drug use potentially limits the ability to generalize the study’s findings to groups underrepresented in the population. Finally, the sample size was small but provided 80% power to detect a change in the incidence of IE of 3.1% or more per year, or a 2.56-fold change during the 31 years of the study. However, this limitation was insurmountable because a 3-decade study of secular trends in endocarditis would be difficult in another population without the unique records-linkage system serving Olmsted County.

In conclusion, our community-based survey revealed no substantial change in the incidence of IE over the past 3 decades. At least in the Olmsted County population, viridans group streptococci continue to outnumber S aureus as the major causative organisms of IE.