Does Asymptomatic Bacteriuria Predict Mortality and Does Antimicrobial Treatment Reduce Mortality in Elderly Ambulatory Women?

Methods

Participant enrollment and the participating institutions have been described previously [10, 11]. Elderly ambulatory residents of the Philadelphia Geriatric Center and of 21 continuing care retirement communities in the greater Philadelphia metropolitan area who gave informed consent were enrolled in this long-term study of urinary tract infection in the elderly. Enrollment continued throughout the course of the study. Philadelphia Geriatric Center houses about 1000 residents who primarily are Jewish; incomes are higher than the maximum Social Security payment; and congregate living is provided either in an apartment house or in a nursing home. In contrast, the continuing care retirement communities are smaller (bed size range, 108 to 675); incomes are higher; residents are primarily not Jewish; and a higher proportion of residents are fully independent.

All female residents were eligible to participate except those with indwelling catheters or those incapable of providing midstream clean-catch urine specimens for culture. Specimens were obtained on enrollment and every 6 months thereafter. The protocol was approved by the appropriate institutional review boards, and informed consent was obtained from the participants or their surrogates.

Table 1 shows the study periods and chronology of important study events. Throughout the study, urine cultures were obtained at about 6-month intervals. An observational study to compare mortality of bacteriuric and nonbacteriuric volunteers regardless of treatment status was begun in January 1983 and ended in February 1992. Initially, residents with asymptomatic bacteriuria were identified and followed, but treatment was not given. However, on 10 October 1983, a controlled clinical trial was begun to evaluate whether antimicrobial therapy for asymptomatic bacteriuria decreased mortality; every bacteriuric study participant identified after this date was enrolled in the trial. Mortality among residents who were treated with antimicrobial agents for asymptomatic bacteriuria each time it was present was compared with the mortality of those who received no therapy for their episodes of bacteriuria. At enrollment, participants were assigned to the treatment group or to the control group based on the last digit of an identification number unrelated to the conduct of the study. Urine cultures were read by personnel blinded to the study group assignment. When asymptomatic bacteriuria was identified, participants with even numbers were given antimicrobial therapy according to a defined protocol (see below); those with odd numbers served as controls. From 10 October 1983 to 10 December 1987, controls were given no therapy. Thereafter, on the advice of external consultants, the protocol was changed so that participants not assigned to the active treatment group were given placebo pills in place of no treatment. The placebo pills were identical in appearance to each of the antimicrobial agents used. Thus, after 10 December 1987, volunteers with asymptomatic bacteriuria were given therapy in either the form of antimicrobics or placebo after a new consent was obtained; the volunteers and clinical personnel did not know study group assignments.

The methods for collecting first-morning urine and for processing the specimens in our research microbiology laboratory have been previously described [10, 11]. Participants were considered to have asymptomatic bacteriuria on a survey if two urine specimens were culture-positive (10⁵ colony-forming units or more per mL of urine) for the same organism within 2 weeks.

From 10 October 1983 through 10 December 1987, residents with asymptomatic bacteriuria who were assigned to receive antimicrobial treatment were given short-course therapy (single dose or 3 days) as follows: trimethoprim, 200 mg in one dose; trimethoprim-sulfamethoxazole, 1 double-strength tablet; cefaclor, 500 mg three times a day for 3 days; amoxicillin, 250 mg three times a day for 3 days; carbenicillin indanyl sodium, four times a day for 3 days; or macrodantin, 100 mg twice a day for 3 days, depending on susceptibility of the infecting organism and history of drug allergy. Participants were considered cured if test-of-cure cultures contained less than 10⁴ colony-forming units/mL of the infecting organism on cultures obtained 5 to 10 days after antimicrobial treatment or placebo or on cultures obtained on the next survey in those receiving no therapy. When positive for the same organism, patients were retreated for 14 days with test-of-cure culture afterward. If the organism differed, reinfection was diagnosed and a single dose or 3-day therapy was used; treatment failures were treated as defined above. Test-of-cure cultures were obtained again after therapy and, if positive for the same organism, participants were treated for 14 days. No treatment was given after failure of a 14-day course or two reinfections after short courses. Controls received no therapy during this period.

After 10 December 1987, culture-positive patients were assigned to antimicrobial treatment or placebo. Single-dose therapy was given with trimethoprim, 200 mg, or norfloxacin, 400 mg, depending on the susceptibility of the organism; the same drugs (trimethoprim, 100 mg twice daily, and norfloxacin, 400 mg twice daily) were used for 14 days of therapy in patients failing single-dose therapy. Single-dose therapy was used for reinfection. The placebo pills and regimen given to a placebo recipient matched the regimen administered to the participant in the active treatment group who was treated most recently (for example, if the active treatment participant received short-course trimethoprim followed by 14 days of trimethoprim, the next placebo participant received short-course trimethoprim placebo followed by 14 days of trimethoprim placebo). Symptomatic infections were managed by the participant's personal physician or by physicians associated with the facility in which the patient lived. Reports were received on an annual or semiannual basis from the participating institutions that detailed changes in their census. All deaths were noted, and registry coordinators reviewed available documents to confirm each death.

After 1 September 1987, detailed functional and mental status assessments were done when persons were newly enrolled into the study or were seen for an annual follow-up visit using techniques previously described [11, 12]. Self-care activities of daily living were assessed by a modification of the Multilevel Assessment Instrument [13], and mental status was assessed using a modified version of the Kahn and Goldfarb questionnaire [14]. A subjective measure of global health status (scale, 1 to 4) was based on responses to the question: “How do you rate your health: excellent (score 1), good, fair, and bad or poor (score 4)?” Diagnoses recorded in the person's medical record were extracted and provided a more objective measure of health status; they were coded according to the ICD-9-CM three-digit codes [15]. The Geriatric Depression Scale [16] was used to assess depressive symptoms, and walking ability was assessed on a scale of 1 (specialized help needed) to 3 (help not needed) [11].

Statistical Analysis

Observational Study

These analyses compared residents with asymptomatic bacteriuria with residents who did not have asymptomatic bacteriuria on any of the urine culture surveys done during the period of their participation. For the purposes of the survival analyses in the observational study, the results of urine cultures were considered a time-dependent covariate. Accordingly, participants were considered in the “ever-positive” group once asymptomatic bacteriuria was identified, and all subsequent time on study was contributed to the group with positive cultures regardless of urine culture results on subsequent surveys. Persons entering the study with a negative urine culture were considered in the “never-positive” group until asymptomatic bacteriuria was identified. Thus, a person who entered the study with negative cultures and who later became culture-positive would have contributed person-time to the follow-up of those in the category of “never positive” as well as those who were “ever positive”. Persons participating in only one survey were excluded from these analyses because the reason for having only one survey might have been related to their health status or risk for death. Moreover, we could not assume for the culture-negative persons, in particular, that they would, in fact, remain culture negative during the subsequent study period. This approach assumes that once bacteriuria is identified that any increased risk for mortality persists regardless of the results of subsequent cultures. Additional analyses were done that included residents who only participated in one survey. Because the conclusions for the primary outcomes were similar, we do not report these results. The demographic and clinical characteristics of the ever-positive group were obtained from the clinical and functional assessment done both at the time of enrollment into the study and at the examination closest to the time that asymptomatic bacteriuria was first identified. The characteristics for those who never had asymptomatic bacteriuria were obtained from the data collected at the time of enrollment into the study or, if enrollment occurred before 1987, at the time of their first interview after September 1987.

Unadjusted comparisons between the ever-positive and never-positive groups were made by first calculating the mortality rate (number of deaths per person-days of follow-up) within each group. The relative rate was computed as the rate in the ever-positive group divided by the rate in the never-positive group. A confidence interval and chi-square test were calculated [17].

To control for differences between the ever- and never-positive groups with respect to baseline covariates, a series of Cox proportional-hazards models was fit using the BMDP (BMDP Statistical Software, Inc., Los Angeles, California) and EGRET (Statistics and Epidemiology Research Corporation, Seattle, Washington) statistical packages. Each of the models used a time-dependent covariate for urinary tract infection status. As above, the effect of this categorization is that persons who eventually became culture positive were counted as “never-positive” until their first positive culture. One model was fit for urinary tract infection status, adjusting for age alone because an a priori decision had been made to include age in the final multivariate model. Univariate models were fit for each of the remaining covariates to try to reduce the number of variables in the subsequent multivariate model. The value of the covariate closest in time to the first positive culture was used for the ever-positive group. Subsequently, a stepwise model was fit using only persons with complete data for all covariates that were significant in univariate models. The two variables (in addition to urinary tract infection status and age) selected by the stepwise procedure were included in a final model that used all individuals in the data set. To include persons with missing values for the covariates, indicator variables for the presence or absence of data for the given covariate were used in the model, taking on a value of 1 for those with valid data for that variable and 0 for those who were missing data for that variable. For each covariate, the interaction between the present or absent variable and the covariate was used in place of the actual covariate. For example, consider the variable for self-rated health status. For those who had valid data recorded for self-rated health, the indicator variable would equal 1, and the interaction term would take on the value of self-rated health. For those missing data for self-rated health, both the indicator variable and the interaction term would have a value of 0. The coefficient for the interaction term is then interpretable as the coefficient for self-rated health for those with valid data for self-rated health and is reported in the Results section as the coefficient for self-rated health. The interpretation of the coefficient for the indicator variable was not considered relevant. A final proportional hazards model was fit in which those who were in the ever-positive group were further subdivided into those who had received active antimicrobial therapy as part of the controlled clinical trial and those who did not receive such therapy.

Controlled Clinical Trial

These analyses considered only those who had (on enrollment) or developed asymptomatic bacteriuria during the course of the study and compared those who received antimicrobial treatment with those who did not. The analyses were done according to the intention-to-treat principle and included patients who participated in only one survey. Baseline clinical and functional characteristics were those obtained closest to the time of a positive urine culture. For the primary analysis of the controlled clinical trial, we used the log-rank procedure [18]. We used a univariate proportional hazards model to estimate the hazard ratio for the treated group compared with the control group with a 95% CI. The data were first analyzed as two separate studies: the first comparing active treatment with no treatment and the second comparing active treatment with placebo. Because the results of these two analyses were very similar with respect to the log-rank test and the hazard ratio, the two studies were combined into an analysis of active treatment compared with a single control group.

We estimated that approximately 540 patients would develop asymptomatic bacteriuria during a 5-year period and agree to participate in the trial. Thus, 270 patients would be assigned to either the group receiving antimicrobial therapy or to the group not receiving antimicrobial therapy. Further, after allowing for 5% loss to follow-up each year, we expected that 140 (58%) deaths would occur in the untreated group. Such a study would have provided 90% power to detect a relative risk (treated compared with untreated) of 0.75, or a 25% reduction in the 5-year risk for mortality caused by treatment.

Results

Observational Study

The baseline characteristics of the women with and without asymptomatic bacteriuria on or after 1 September 1987, the time when detailed clinical and functional data became available, are shown in Table 2. Of 1750 (330 culture-positive) eligible enrollees (see Table 1), 1491 [318 culture-positive] participated in more than one survey and were included in this analysis; 412 of the 1491 were enrolled before September 1987 and were being followed. As a group, those who ever had asymptomatic bacteriuria were older and sicker than those who had never had asymptomatic bacteriuria. With the exception of the number of hospitalizations, the findings for the ever-positive group differed significantly from those for the never-positive study participants in a direction indicating that they were less well.

The death rate in the never-positive group was 10.1 per 100 000 resident-days (Table 3). In contrast, the death rate in the ever-positive group was higher, 18.7 per 100 000 resident days. The unadjusted relative rate was 1.85 (CI, 1.33 to 2.58).

Using a Cox proportional-hazards model (Table 4), age at entry was associated with an increased mortality rate (hazard ratio per unit change in age, 1.07; CI, 1.04 to 1.10), and a strong relation was found between self-rated health (hazard ratio per unit change in self-rated health, 1.62; CI, 1.31 to 2.00) and subsequent mortality. No relation was found between ever having had asymptomatic bacteriuria and death after adjusting for covariates (hazard ratio, 1.10; CI, 0.78 to 1.55).

We also analyzed the data considering whether or not the participant in the observational study was assigned to receive treatment. In this analysis, the death rate among participants in the never-positive group (122 of 1173; 10.4%) was compared with that among those who were ever positive but received no treatment (18 of 134; 13.4%) and with those who were ever positive and received antimicrobial therapy (31 of 184; 16.8%). The comparison across all three groups was not statistically significant using a chi-square test (P > 0.2). Additional analyses adjusting for age and other covariates were done using time-dependent covariate proportional hazards models (Table 5). As noted previously, age at baseline and self-rated health were strongly associated with mortality, but the number of drugs was no longer associated with death. The difference between those assigned to receive treatment and those assigned to receive no treatment was found not to be significant. In addition to the analysis of patients who had detailed functional assessments after 1 September 1987 as above, additional analyses (not shown here) were done that included participants enrolled before September as well as those enrolled after that date (2162 volunteers; 1863 participants on two or more surveys); the results did not change. Analyses were also repeated using covariates defined at study entry for the ever-positive group, again with no change in results.

With the usual reservations about documents describing causes of death in mind, it did not appear that urinary tract infections or their complications were major causative factors.

Controlled Clinical Trial

No differences in baseline characteristics were detected between the 192 treated participants and 166 controls (Table 6). Thirty patients (18.1%) died in the treated group and the death rate was 13.8 per 100 000 resident-days; this compares with 39 deaths (20.3%) and a death rate of 15.1 per 100 000 resident-days in the control group; the relative rate was 0.92 (CI, 0.50 to 1.47). Survival curves for the treated and control groups are shown in Figure 1; the differences were not significant (P > 0.2). When data before and after 1987 were analyzed separately, the results were comparable. When the independent effect of self-rated health on survival was assessed using combined data from before and after 1987, self-rated health remained a significant predictor of mortality (hazard ratio, 2.00; CI, 1.43 to 2.78; P < 0.001). Adjustment for age and self-rated health, using a proportional hazards model, did not affect the comparison between the treated and control groups. The overall cure rates during the placebo-controlled portion of the trial were 82.9% in those given antimicrobial agents and 15.6% in those not given antimicrobics in an intention-to-treat analysis.

View larger version:

• In this window
• In a new window

Figure 1. ( > 0.2). Mortality in patients treated with antimicrobial agents and controls.P

Discussion

A relation between asymptomatic bacteriuria and increased mortality has been reported in several studies and in several different clinical settings. In 1973, Sourander and Kasanen [1] described a population-based study of asymptomatic bacteriuria in those older than 65 years in Turku, Finland. They found a significantly higher death rate in the subgroup of women from 75 to 79 years with bacteriuria compared with those of similar ages without bacteriuria. In 1981, Dontas and colleagues [2] reported on the relation between asymptomatic bacteriuria and death among 342 healthy residents of a home for the aged. The median survival of men (33 months) and women (34 months) with asymptomatic bacteriuria was significantly shorter than that of men (53 months) and women (75 months) without bacteriuria in those aged 70 to 79 years. The absence of significant differences in several covariates (age distribution, blood pressure, hematocrit, smoking habits, cholesterol levels, or myocardial changes) between those with and without bacteriuria in the latter study supported a conclusion that “bacteriuria in old age is associated with a reduction in survival of 30 to 50 per cent.” This particular study was considered important and stimulated additional studies, including our own.

A relation between urinary tract infection and mortality has been identified in other clinical settings. Such studies support the plausibility of a relation between urinary tract infection and death. In a community study in Wales and Jamaica, Evans and colleagues [3] found a relation between bacteriuria and mortality. In this study (three urine culture surveys during 13 years) of women aged 15 to 84 years, they found increased mortality in bacteriuric participants even after adjusting for age and weight. In a study of hospitalized patients, Platt and colleagues [4] found nearly a threefold increase in mortality among patients with urinary catheters who developed urinary tract infection compared with those without catheters and without infection. Although several variables were identified as risk factors for death in univariate and multivariate analyses, precise reasons explaining the findings were not identified. Finally, Kunin and colleagues [19] reported that urinary catheter use had an independent effect on mortality (and morbidity) among elderly nursing home patients after adjustment for several cofactors; however, they did not specifically assess urinary tract infection.

Others have been unable to confirm a relation between asymptomatic bacteriuria and death in the elderly. In 1986, Heinamaki and colleagues [5] described results of an observational 5-year study done in 83% of the 674 residents of Tampere, Finland, who were 85 years or older. In that study, survival was not related to bacteriuria as determined by a single urine culture on entry to the study. In that same year, Nordenstamm and colleagues [6] reported no relation in women between bacteriuria and death during 9 years. Bacteriuria and survival were related in men in an adjusted analysis. However, after adjusting for occurrence of cancer, mortality among bacteriuric men was not found to be greater than among controls. An important feature of this study was that age was rigorously controlled—all participants were 70 years old and the standard deviation was about 2 months. Nicolle and colleagues also failed to identify a relation in a study in women [7] and in a second study in men [8]. In an 11-year study of 408 persons in a residential home, Dontas and colleagues [9] failed to find an association between persistent bacteriuria and increased mortality, but decreased mobility, smoking, and hematocrit at entry were associated with increased mortality. In one of our earlier studies, we also found a relation between mobility and mortality [11].

The studies evaluating a relation between asymptomatic bacteriuria and mortality in the elderly have been criticized. Some studies classified the participants on the basis of the results of only one urine culture [1, 5], but in the absence of symptoms two cultures positive for the same organism are usually required to reliably identify bacteriuric persons. Some studies stratified participants as bacteriuric or not based on urine culture results obtained at enrollment and did not allow for identification of persons in the culture-negative group who might have been identified as having developed bacteriuria during the course of the study. It is certain, particularly in long-term studies, that an appreciable proportion of culture-negative persons would have been identified as having developed asymptomatic bacteriuria had additional cultures been obtained [10]. Finally, the possibility has been raised that urinary tract infection in itself was not related to death but merely a marker for illness in sicker persons.

Our study was designed with these issues in mind. Participants were considered culture positive only if two cultures were positive for the same organism over 2 weeks. Serial cross-sectional surveys were done to identify participants who developed bacteriuria during the course of the study. The optimum interval for such surveys is not known, but had cultures been obtained at even more frequent intervals, probably more culture-positive participants would have been identified [10]. Importantly, our method of analysis allowed a person to contribute time to the analysis of the culture-negative group as well as time to the culture-positive group. Finally, both medical and functional status were carefully assessed.

The results of the observational study comparing culture-positive and culture-negative persons clearly indicate that culture-positive persons were, as a group, older and sicker. For each of the indicators of illness—total number of diagnoses, total drugs, hospitalizations, activities of daily living, global depression score, and so forth—the differences between the culture-positive and -negative groups were statistically significant. Unadjusted death rates between the culture-positive (19%) and -negative groups (10%) differed significantly; however, after adjustments for important covariates were made, the differences were not significant. In analyses using the Cox proportional-hazards model, age at entry and self-rated health were significantly associated with increased mortality. Additional analyses by specific treatment group also failed to confirm an association between bacteriuria and death. Within the limitations of the study, our results do not support an association between asymptomatic bacteriuria and mortality in elderly ambulatory women.

The results of our controlled clinical trial comparing mortality in treated and untreated patients with asymptomatic bacteriuria also failed to provide evidence of such an association. The mortality among treated and untreated participants was 13.8 per 100 000 and 15.1 per 100 000, respectively; the rate ratio was 0.92 (CI, 0.57 to 1.47). Fewer events occurred than were anticipated, decreasing power below that projected; this effect can be seen by examining the width of the confidence intervals. It is important to note, however, that the point estimates for the rate ratio in both the observational study and the controlled clinical trial approach 1.

Others have examined the effect of antimicrobial therapy on mortality in elderly bacteriuric patients. Nicolle and colleagues [8] identified institutionalized elderly men and randomly assigned them to therapy or no therapy. Despite therapy, participants did not remain abacteriuric for long periods, and no effect on mortality was identified. In a previous report [20] and in this study, we noted an excellent bacteriologic response to antimicrobial therapy among ambulatory elderly women. Mortality, however, was unaffected by treatment.

In addition, an independent effect of self-rated health on risk for death was seen in this study after controlling for other more objective measures of health status. This finding is consistent with the results of several large epidemiologic studies [21-25]. Although the evidence supporting the importance of global self-ratings of health for mortality is consistent across studies, the mechanism or mechanisms by which self-rated health is related to the risk for dying are not clear. Is self-rated health a reflection of a prescient understanding of biological and physiologic changes that lead one to perceive one's health more correctly than inferred by more objective indicators? Does poor self-rated health merely reflect an individual's recognition of previous or present behaviors that could adversely affect their health such as smoking or alcohol abuse? Alternatively, are positive ratings protective because optimistic feelings are in themselves protective? Finally, is one's self-health rating really a proxy for the individual's emotional rather than physical status? The data in this study, as well as in the literature, are insufficient to evaluate these hypotheses adequately.

We failed to confirm the purported relation between asymptomatic bacteriuria and increased mortality among ambulatory elderly women. We did not find an increase in mortality among bacteriuric volunteers in our longitudinal observational study, and we could not show a survival benefit from antimicrobial therapy for asymptomatic bacteriuria in a controlled clinical trial. Our results suggest that mortality in elderly women with asymptomatic bacteriuria cannot be reduced by antimicrobial therapy. Within the limitations of this study, our results provide evidence that asymptomatic bacteriuria in the elderly should not be treated if the goal is to decrease mortality. It should be noted that we did not study morbidity from urinary tract infection, and our results do not address potential reductions in morbidity by treatment. In this regard, we [26] and others [7, 27-31] have advised that screening for asymptomatic bacteriuria and treatment with antimicrobial therapy is generally not warranted. Lastly, we confirmed that a simple subjective measure, self-rated health, can be used to differentiate a group with increased mortality.