Dental and Cardiac Risk Factors for Infective Endocarditis: A Population-Based, Case-Control Study

Methods

Participants

From August 1988 to November 1990, we maintained surveillance for infective endocarditis in 54 hospitals of the Delaware Valley Case-Control Network, a population-based network of hospitals in the eight counties that constitute the Philadelphia Metropolitan Statistical Area and the county of New Castle, Delaware. Patients with a putative diagnosis of infective endocarditis were identified by hospital personnel and were reported to study nurses, who also actively sought cases. To assess the completeness of ascertainment, five high-yield hospitals and three low-yield hospitals were asked to list all patients discharged with a diagnosis of endocarditis over 3 months. These lists were compared with those obtained from our surveillance; charts were reviewed when differences were identified.

We obtained informed consent from physicians and case-patients, then used structured forms to abstract medical records, including echocardiographic reports and hospital laboratory information on the infecting organism. We deleted information on purported risk factors for infective endocarditis and submitted these records for review by three of the authors, who are consultants in infectious diseases recognized for their expertise in infective endocarditis [5, 6]. These experts used their own global clinical judgment to classify potential cases as definite, probable, or possible cases or probable noncases. Agreement of two of the three reviewers was required to make the determination of a case or a noncase [6].

One control from the community was selected for each case-patient by using a modification of the Waksberg random-digit dialing method [7]. Controls and case-patients were matched for age (in 5-year strata), sex, and neighborhood of residence (by using area code, telephone exchange, and the first subsequent digit of the case-patient's telephone number). We excluded from these analyses patients with infective endocarditis who were younger than 18 years of age, intravenous drug users, and patients who developed endocarditis in the hospital.

This study received separate institutional review board approval at the University of Pennsylvania and all 54 participating hospitals.

Data Collection

Information was obtained from case-patients by conducting a structured telephone interview after hospital discharge. The date of hospital admission served as the “study date” for case-patients; for controls, the date of the telephone interview was used. Telephone interviewers collected information on demographic characteristics; diagnostic and therapeutic medical and dental procedures in the year before the study date; potential host risk factors, including preexisting cardiac lesions, preexisting local infection, risk factors for oral or dental disease, diabetes mellitus, immune deficiencies, family history of endocarditis, alcoholism, malignant conditions, and autoimmune disease; previous antibiotic use; and other recent illnesses. For each host risk factor, we requested the date of diagnosis, diagnostic method (for example, echocardiography for mitral valve prolapse), and the name of the practitioner who made the diagnosis. For each medical and dental procedure, we sought information about the procedure, the month and year in which it was performed, and the practitioner. We requested medical and dental records describing procedures and validating individual diagnoses.

Study Variables

Case-patients were considered infected with dental flora if the organism found was viridans streptococci; nutritionally variant streptococci; Actinobacillus species; Cardiobacterium hominis; anaerobes; α-hemolytic streptococci (not group D); unspecified streptococci; or Haemophilus, Eikenella, Kingella, or Neisseria species.

Because this study focused on indications for antibiotic prophylaxis, we examined host characteristics reported by patients as the primary risk factor variables, reflecting the information that would be available to a practitioner about to perform a procedure for which prophylaxis might be indicated. A variable called “any valvular heart abnormality” was defined as the presence of any of the following self-reported, preexisting conditions: mitral valve prolapse, congenital heart disease, history of rheumatic fever with heart involvement, prosthetic heart valve, previous episode of endocarditis, or other valvular heart disease.

Dental visit information was obtained solely from dental records. Dental hygiene care was defined as preventive oral health services and therapeutic services, including coronal scaling and polishing. Consistent with AHA guidelines, invasive dental procedures were defined to include dental hygiene care, extractions, periodontal treatment (including scaling and root planing), endodontic treatment, mouth or gingival surgery, and treatment of tooth abscess. Noninvasive dental procedures were simple restorations, prosthetic and restorative dentistry, fluoride treatment, and other procedures (prosthetic services, including adjustments and suture removal). Unless otherwise specified, dental treatment refers to all dental treatment and is not limited to invasive procedures.

Initial analyses focused on dental procedures performed at any time in the 3 months before the study date. Analyses were then narrowed to 2 months and 1 month before the study date. Time frames are approximate because the onset of infective endocarditis is often difficult to determine with certainty. We therefore chose the date of hospital admission as the study date, collected procedural data based on month rather than on a specific date, and calculated time frames under the assumption that procedures were performed on the 15th of the month.

Statistical Analysis

Frequencies and cross-tabulations between case–control status and potential risk factors were obtained. Conditional logistic regression was used to determine the independent effects of the various potential risk factors and the possibility of any interactions among factors [8]. Variables were included in multiple regression models if they were significant (P < 0.2) in unadjusted (matched) analyses (such as kidney disease and diabetes), if their inclusion had a substantial effect (>15% change) on coefficients for variables already in the model (such as insurance status) [9], or if they were strongly suspected a priori of being confounders (such as ethnicity).

For analyses specific to participants with known cardiac valvular abnormalities, odds ratios and CIs were calculated from a model that included main effects for cardiac valvular abnormalities and dental treatment and the interaction between those variables. The odds ratio for various dental therapy variables among participants with cardiac valvular abnormalities was estimated by combining coefficients for the dental therapy variable and the interaction term. The CI for this combination of coefficients was estimated by using the appropriate variance and covariance terms [8]. With the interaction terms, participants with and those without valvular abnormalities were included in these analyses. Exact odds ratios and CIs, stratified on the matching variables, were calculated when data were too sparse for conditional logistic regression [10].

We used SAS statistical software (SAS Institute Inc, Cary, North Carolina) for data management and to obtain frequencies and cross-tabulations. We used EGRET (Epidemiological Graphics, Estimation and Testing software, version 0.25.1, Cytel Software Corp., Cambridge, Massachusetts) for conditional logistic regressions and exact analyses. All CIs are 95%, and all P values are two-sided. The sample size for the study was chosen so that by assuming an α level of 0.05 (two-sided) and a power of 80%, we would be able to detect associations with an odds ratio of 2.0 or more for risk factors with a prevalence between 0.1 and 0.8.

Results

Participants

We identified and recruited 416 potential case-patients (Figure 1). Our assessment process confirmed that more than 90% of true cases of infective endocarditis had been identified. The expert panel judged 379 patients to have definite, probable, or possible infective endocarditis; 37 (9%) were judged to be probable noncases and were excluded. Agreement among judgments was high, ranging from 92% to 96% [6].

Of these 379 patients, 287 had community-acquired infective endocarditis not associated with intravenous drug use (248 on native valves and 39 on prosthetic valves), 27 had nosocomial infective endocarditis (18 on native valves and 9 on prosthetic valves), and 65 were intravenous drug users. Of the 287 patients, 273 (95%) completed the interview; these case-patients are the focus of our study. Seventy-five other patients who were identified as possibly having community-acquired disease and were not intravenous drug users were not recruited because consent was not obtained. Next-of-kin or other surrogate respondents completed the interviews for 25% of case-patients. Repeated analyses omitting surrogate respondents did not substantively change the results.

Screening interviews identified community residents who were matches for case-patients; interviews were completed with 58% of the matches. Case-patients for whom more than one matched community resident was invited to participate before a completed interview was obtained did not differ by age or sex from case-patients for whom the interview was completed by the first identified community match.

Of the 1381 medical and dental records requested, 1265 (92%) were received. Records were requested for 493 participants and were received for 480 (97%); proportions were similar for case-patients and controls. Data on exposures from medical records agreed with those from the interviews for more than 95% of controls. For case-patients, agreement between interview responses and medical records was more than 90% for most variables and was often more than 95%; for next of kin, agreement was more than 80%; and for other surrogates, it was more than 80% and often more than 90%. Preexisting valvular heart disease was confirmed by medical records in 98 (94%) of the 104 case-patients and 11 (65%) of the 17 controls identified by interview as having this condition.

Infecting Organisms

Of the 287 patients with community-acquired infective endocarditis not associated with intravenous drug use, 272 (95%) had multiple positive blood cultures (Table 1). Of the 15 case-patients with negative cultures, 12 received antibiotics before admission, 4 had histopathologic evidence of endocarditis, and 10 had echocardiographic evidence of valve vegetation. One case-patient had an inconclusive transthoracic echocardiogram but had fever for 6 days and antibiotic therapy for 2 days before blood culture was performed; this may account for the negative results.

Demographic Characteristics

Case-patients and controls were similar with regard to age and sex (matching variables), ethnicity, education, occupation, and dental insurance status. Case-patients and controls ranged from 18 to 98 years of age (mean age, 59.1 ± 17.1 and 59.1 ± 17.0, respectively). Case-patients were more likely than controls to have their care paid for by a government program, such as veterans' benefits or Medicaid (P = 0.001).

Dental Risk Factors

Dental Procedures

In the 2 months before the study date, 16.8% of case-patients and 14.3% of controls had dental treatment; these proportions increased to 23% of case-patients and 23% of controls in the 3 months before hospital admission (Table 2). No individual dental procedure was significantly associated with infective endocarditis (Table 3), except tooth extraction in the 2 months before hospital admission (P = 0.03). However, because tooth extractions were uncommon (performed in 6 case-patients and no controls), inclusion of this variable in the full model was precluded.

Risk in Persons Infected with Dental Flora

When we restricted analyses to case-patients infected with dental flora and their matched controls, results were still negative (Table 2). The adjusted odds ratio for dental procedures within 2 months of hospital admission was higher than that within 1 month, but the lowest unadjusted and adjusted odds ratios were found at 3 months. Case-patients were more likely than controls to have had dental treatment in the unadjusted analysis at 2 months, but these results were of borderline significance (odds ratio, 2.2 [95% CI, 1.01 to 4.9]), and the association disappeared in the adjusted analysis (odds ratio, 1.5 [CI, 0.3 to 7.0]).

Cardiac Risk Factors

A patient-reported history of any cardiac valvular abnormality was highly associated with infective endocarditis (Table 4) (adjusted odds ratio, 16.7 [CI, 7.4 to 37.4] adjusted for socioeconomic status variables [ethnic group, education, occupation, health insurance status, and dental insurance status], diabetes mellitus, and severe kidney disease). Case-patients were substantially more likely than controls to report having previously known mitral valve prolapse, history of congenital heart disease, rheumatic fever, cardiac valvular surgery, a previous episode of endocarditis, other valvular heart disease, or heart murmur without other known cardiac abnormalities.

Risk of Dental Treatment in Persons with Known Cardiac Valvular Abnormalities

Among case-patients and controls with known cardiac abnormalities-the target of antibiotic prophylaxis (Table 2)-the risk for infective endocarditis was not increased by dental treatment. Of 104 case-patients with a known cardiac valve abnormality, 21 (20.2%) had had dental treatment within 2 months of the study date; 4 of 17 (23.5%) controls had had a similar exposure (adjusted odds ratio, 0.3 [CI, 0.05 to 1.4]). Within a 3-month period, 29 (27.9%) of the case-patients and 6 (35.3%) of the controls had had dental treatment (adjusted odds ratio, 0.2 [CI, 0.04 to 0.7]).

The distribution of the “incubation periods” of the 29 case-patients with cardiac valvular abnormalities and dental treatment in the previous 3 months was examined, and no clear pattern emerged. Among 18 case-patients who could identify the date of symptom onset, 4 had dental treatment after the onset of symptoms, 7 had treatment in the first month before onset, 4 had treatment in the second month before onset, and 3 had treatment in the third month before onset. These results again suggest that no relation exists between dental treatment and illness.

Of note, the proportion of case-patients who had both a known cardiac valvular abnormality and dental treatment was small (10.6% [29 case-patients] who had had dental treatment in the previous 3 months), indicating that the proportion of infective endocarditis cases even theoretically preventable by prophylaxis is small. Only 12 case-patients (4.4%) had had dental therapy 1 month before the study date.

Risk in Persons with Known Cardiac Valvular Abnormalities and Infection with Dental Flora

Among 56 case-patients with known cardiac abnormalities who were infected with dental flora and their controls (Table 2), we found no significantly increased risk from dental treatment (although these numbers were small).

Antibiotic Prophylaxis

Only 6 case-patients (2.2%) and 2 controls (0.7%) received antibiotic prophylaxis within 1 month before the study date. The numbers for 2 and 3 months, respectively, were 14 (5.1%) and 24 (8.8%) for case-patients and 3 (1.1%) and 3 (1.1%) for controls. Adjustment for this use in the multivariate analyses (equivalent to restricting the analyses of dental procedures to participants who did not have prophylaxis) did not substantively change the results (Table 2).

We also performed an analysis to assess the value of antibiotic prophylaxis, but sample sizes were too small for the analysis to be conclusive. For participants with cardiac valvular abnormalities who had dental therapy, the risk for infective endocarditis remained the same regardless of use of prophylactic antibiotics (P > 0.2). The unadjusted odds ratio for dental therapy within 3 months (compared with no dental therapy) was 0.5 (CI, 0.01 to 9.6) with antibiotic prophylaxis and 0.3 (CI, 0.01 to 4.2) without antibiotic prophylaxis. Furthermore, among case-patients with known cardiac abnormalities who were infected with dental flora, 6 of 8 who had had an invasive procedure within 1 month, 9 of 13 who had had an invasive procedure within 2 months, and 12 of 16 who had had an invasive procedure within 3 months received antibiotic prophylaxis. Review of dental records indicated that prescription of antibiotics was consistent with AHA guidelines.

Discussion

In this large, population-based, case–control study of infective endocarditis, case-patients were no more likely than controls to have received dental therapy. Furthermore, case-patients were substantially more likely than controls to have various cardiac lesions, but the presence of these abnormalities did not result in an increased risk for infective endocarditis from dental procedures. Few participants received antibiotic prophylaxis; therefore, this did not mask the effect of dental procedures.

A possible risk associated with dental extractions within 2 months was identified, but this finding was based on only 6 case-patients, could not be adjusted for possible confounding, and was not evident in the subgroup of participants with valvular abnormalities, the true target of prophylaxis. Thus, this may be a chance finding in the face of the large number of statistical tests performed. A second positive finding regarding risk among participants infected with dental flora (odds ratio, 2.2 [CI, 1.01 to 4.9]), also within 2 months, diminished with adjustment (odds ratio, 1.5 [CI, 0.3 to 7.0]). Although this subgroup is of biological interest, it is not the target of prophylaxis per se because persons who will have endocarditis from dental flora compared with those who have endocarditis from nondental flora cannot be identified in advance. Furthermore, this finding was not confirmed among participants with dental flora who had cardiac abnormalities.

A recent review claimed that “about 15% of patients with infective endocarditis acquire the disease in the process of dental manipulation” [3]. In our study, 23.1% of case-patients had had dental therapy within the previous 3 months and 9.9% had had invasive dental procedures. However, our study is the first to choose a random sample of community residents to document the frequency of dental treatment. Of note, we found no increased prevalence of dental treatment among case-patients compared with controls. In fact, transient bacteremia is documented as part of everyday life, coincidental with tooth brushing, chewing food after tooth brushing, and, presumably, bowel movements. Until now, it was not clear whether the cumulative risk of endocarditis from daily low-grade bacteremia was exceeded by that associated with the occasional dental procedure [11]. Moreover, among patients with infective endocarditis, those with and those without previous dental extractions have the same frequency of infection with viridans streptococci [12]. Inasmuch as dental treatment provides the presumed cause of this infection, the significance of dental treatment as a risk factor for endocarditis is questionable.

To our knowledge, only one controlled study of dental risk factors for infective endocarditis has been done: a case–control study of 171 (of 415) case-patients with infective endocarditis and matched controls from medical and cardiology wards [4]. No increased risk associated with dental procedures in the preceding 90 days was seen, overall (odds ratio, 1.2 [CI, 0.7 to 2.1]) or with extractions (odds ratio, 0.7 [CI, 0.3 to 1.8]). A borderline increased risk was seen only with endodontic treatment and dental scaling; the latter was significantly associated (odds ratio, 5.3; P = 0.025) with viridans streptococcal endocarditis. This single subgroup finding might have resulted from multiple statistical tests and was not confirmed here.

We are not aware of any previous controlled human studies that quantify the risk for infective endocarditis associated with cardiac abnormalities, with the exception of mitral valve prolapse. The reported relative risk for endocarditis from mitral valve prolapse has ranged from 1.0 to 27.4 [13-16], depending on the inclusion criteria used. We deliberately limited mitral valve prolapse to that already identified by a physician and known to the patient, because these patients are currently candidates for prophylaxis. In addition, because the diagnosis of mitral valve prolapse was based on echocardiography and other diagnostic tests performed in numerous facilities over the course of years, we did not believe that valid differentiation could be made about regurgitation. The risk for mitral valve prolapse restricted to regurgitation is undoubtedly higher [13, 15, 17, 18].

Infective endocarditis is even theoretically preventable in a small proportion of patients: In our study, only 10.6% of case-patients were eligible for prophylaxis and only 23% of all case-patients had dental treatment. A case–control study from the Netherlands [19] cites a proportion of 11.0%. On the basis of figures from earlier literature, Kaye and Abrutyn [11] calculated that prophylaxis could prevent fewer than 10% of cases in patients with dental flora, assuming 100% effectiveness. Our data show that 38.8% of case-patients were infected with dental flora, the prophylaxis target. At a baseline incidence of 5.14 community non-intravenous drug user cases per 100 000 person-years in the general population [2], even at 100% effectiveness, prophylaxis would reduce incidence by only 2.0 cases per 1 000 000 person-years.

Antibiotic prophylaxis in at-risk patients has been accepted as reasonable practice for four decades [1, 3], although data confirming its effectiveness are lacking. Three case–control studies of antibiotic prophylaxis have been performed. In one, investigators assembled 8 cases over 7 years from two hospitals and relied on patient recall of antibiotic prophylactic use; reported efficacy was 91% [20]. In the second study, 48 cases from a nationwide Dutch population were collected over 2 years; antibiotic prophylactic efficacy was not confirmed [19]. Finally, 171 of 415 case-patients described above showed overall benefit from antibiotic prophylaxis compared with hospital controls but showed less benefit with viridans streptococcal or culture-negative infective endocarditis (that is, the patients who should have benefited the most) [4].

Prophylaxis fails even with antibiotic regimens recommended by the AHA/American Dental Association [21-24], as we observed. However, with little convincing evidence from human studies, expert groups continue to specify antibiotic prophylactic regimens [1, 18]. Against low incidence and questionable efficacy, one must balance the rare but real risk for adverse reactions, including anaphylaxis, and the possible occurrence of drug-resistant organisms [25]. In a 1979 analysis, the risks associated with penicillin prophylaxis seemed to outweigh its benefits, although prophylaxis with erythromycin, considered less effective but also less risky, could be substituted [26]. A recent cost-effectiveness analysis concluded that prophylaxis was warranted, but only with dental extractions [27]. However, this conclusion requires the unjustifiable assumptions that 15% of cases of endocarditis are directly attributable to (as opposed to preceded by) dental procedures and that 95% of those cases are attributable to extractions.

To protect against selection bias, we chose a population-based design and enrolled more than 90% of all patients with infective endocarditis in the Delaware Valley. Random-digit dialing was used to enroll controls. To assure a valid definition of infective endocarditis, we chose recognized authorities to classify cases, requesting their best clinical judgment after consideration of clinical, echocardiographic, and microbiological findings. The conventional von Reyn criteria [28] were developed before the widespread use of echocardiography. The Duke criteria [29], unavailable when this study began, include risk factors to which the experts were blinded. Instead, experts were provided information on the von Reyn criteria and echocardiographic results but were blinded to risk factors. The diagnoses of the experts agreed closely with each other and with those resulting from use of the Duke criteria [5, 6].

It seems unlikely that information bias played a role in our study. Interviewers and medical record abstracters were not blinded to study group but were extensively trained in good interviewing and abstracting techniques. Explicit quality-control measures and analyses were imposed, and data were collected by using standardized forms. Host risk-factor data were the same data that clinicians would have when deciding whether to prescribe antibiotic prophylaxis. Although surrogate interviews were obtained for some cases, analyses were repeated after exclusion of these interviews and the results did not change substantively.

We sought to quantitate the risks associated with specific subgroups and specific dental procedures, but the numbers of participants in some analyses were small. Our ability to exclude these associations was therefore weak, as reflected by the wide CIs. However, these small numbers also indicate a low frequency of endocarditis with recent exposure to these procedures, despite the extensive population base and the volume of dental therapy provided in the Delaware Valley over 2 years. If any associations exist, too few cases result to justify the expense and risk of widespread antibiotic prophylaxis.

In conclusion, we did not find dental procedures to be a risk factor for endocarditis, even in patients with underlying cardiac valvular abnormalities. We also confirmed the importance of preexisting cardiac abnormalities as principal risk factors. Only a few cases of infective endocarditis could be prevented by antibiotic prophylaxis for dental procedures, even if 100% effectiveness were assumed. We conclude that the policies for administering antibiotic prophylaxis to patients with cardiac abnormalities undergoing dental treatment should be reconsidered.

The authors thank the patients and the community residents and their physicians and dentists who complied with our requests for information and records. In addition, they acknowledge the vital roles that Margaret Dunn, RN, MA, Connie Marrone, RN, and the interviewers played in this study and thank Lynn Rotoli for her expertise in the preparation of the manuscript.

Dr. Abrutyn: Department of Medicine and School of Public Health, Allegheny University of the Health Sciences/Hahnemann, Mail Stop 487, Broad & Vine Streets, Philadelphia, PA 19102-1192. Dr. Berlin: Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania, 611 Blockley Hall, 423 Guardian Drive, Philadelphia, PA 19104-6021.

Ms. Kinman: Center for Clinical Epidemiology and Biostatistics, University of Pennsylvania, 826 Blockley Hall, 423 Guardian Drive, Philadelphia, PA 19104-6021.

Dr. Feldman: Veterans Administration Medical Center, 307B, University & Woodland Avenues, Philadelphia, PA 19104.

Dr. Stolley: Department of Epidemiology and Preventive Medicine, University of Maryland, 660 West Redmond Street, Suite 109, Baltimore, MD 21201.

Drs. Levison, Korzeniowski, and Kaye: Department of Medicine and School of Public Health, Allegheny University of the Health Sciences/MCP, 3300 Henry Avenue, Philadelphia, PA 19129.