Published ahead of print on December 23, 2003, doi:10.1164/rccm.200310-1448OC
© 2004 American Thoracic Society
Medium-Term Survival after Hospitalization with Community-Acquired PneumoniaDepartment of Medicine, University of Western Australia, and Department of Respiratory Medicine, Royal Perth Hospital, Perth, Western Australia, Australia; and Physicians Research Network, Methodist Le Bonheur Healthcare, and Clinical Research, Methodist Healthcare, Memphis, Tennessee Correspondence and requests for reprints should be addressed to Grant W. Waterer, M.B.B.S., F.R.A.C.P., School of Medicine and Pharmacology, University of Western Australia, and Department of Respiratory Medicine, Royal Perth Hospital, G.P.O. Box X2213, Perth 6847, Western Australia, Australia. E-mail: waterer{at}cyllene.uwa.edu.au
An episode of community-acquired pneumonia (CAP) has been suggested to predict greater than expected mortality after discharge from hospital. We ascertained the survival status as of December 2002 of a cohort of patients with CAP identified prospectively between November 1998 and June 2001. Cox regression analysis was used to examine the impact of demographic factors, comorbid illnesses, and CAP severity on subsequent mortality. Of 378 CAP survivors we ascertained the survival status of 366 (96.9%), 125 (34.1%) of whom had died. The mean length of follow-up was 1,058 days (range, 602–1,500 days). Independent predictors of mortality were increasing age (p < 0.001), comorbid cerebrovascular (p = 0.002) and cardiovascular (p = 0.023) disease, an altered mental state (p < 0.001), a hematocrit of less than 35% (p = 0.035), and increasing blood glucose level (p = 0.025). In 41- to 80-year-olds without significant comorbidities there was a trend to greater than expected mortality. In conclusion, an episode of CAP in young adults without significant comorbid illnesses does not appear to be an adverse prognostic marker of medium-term survival. The trend to greater than expected mortality in patients over 40 years of age needs further study and physicians should be particularly alert for underlying life-limiting disease processes in patients presenting with acute confusion or a hematocrit of less than 35%.
Key Words: hospitalization • pneumonia, survival long-term Community-acquired pneumonia (CAP) is both common and a major cause of morbidity and mortality in most Western countries, including the United States (1). Risk factors for acquiring CAP include increasing age and comorbid illnesses, such as cardiac failure, diabetes, neoplasia, and chronic obstructive pulmonary disease (2, 3). When a predisposing factor cannot be identified, potential explanations include a particularly virulent pathogen or an underlying genetic predisposition. Some patients have no identifiable cause despite exhaustive investigation. Several studies have shown that patients surviving an intensive care unit admission have a subsequent mortality rate substantially greater than that of age-matched control subjects (4–6). The greater mortality rate is largely explained by the more common occurrence of comorbid illnesses in intensive care unit subjects (4–6). Brancati and coworkers followed a cohort of patients surviving an episode of pneumonia requiring hospitalization (7). The only independent predictors of 2-year mortality were comorbid illnesses and a hematocrit of less than 35%. As age was not an independent predictor of subsequent mortality, this raises the possibility that for some patients an episode of pneumonia could be a sentinel event for an underlying life-limiting disease. Even if the findings of Brancati and colleagues cannot be confirmed, identification of predictors of subsequent mortality after discharge with CAP may have important implications for both patients and their treating physicians. We therefore ascertained the survival status up to 4 years after discharge from hospital of a prospectively collected cohort of patients with CAP to study the risk factors for subsequent mortality. Some of the results of this study were reported in abstract form at the American College of Chest Physicians Annual Scientific Meeting in 2003 (8).
Study Design A prospective cohort of patients admitted to the Methodist Healthcare Memphis Hospitals (Memphis, Tennessee) with CAP was recruited between November 1, 1998 and June 30, 2001. An attempt was made to enroll all patients admitted with CAP during this period; however, subjects were enrolled only after written informed consent was obtained. The Methodist Healthcare Institutional Review Board approved the study. Survival was ascertained for subjects as of December 31, 2002, using social security number-linked death records, review of hospital and outpatient pharmacy records, contact with all known treating physicians, and postal contact at the last known home address.
Inclusion Criteria
Exclusion Criteria
Data Collection
Definitions
Statistical Analysis
Of an initial cohort of 404 subjects with CAP, 378 (93.6%) survived to hospital discharge. We were able to ascertain the survival status of 366 (96.9%) of the 378 survivors. These 366 subjects had a mean age of 58.1 years (range, 18–99 years): 54.9% were female, 62.0% were African American, and 38.0% were white. Death occurred in 125 patients (34.1%) after discharge from hospital. The mean length of follow-up was 901 days (range, 5–1,500 days) for all patients and 1,058 days (range, 602–1,500 days) for all survivors. For nonsurvivors the mean time to death was 435 days (range, 15–1,310 days). A Kaplan–Meier plot of mortality by age groups is shown in Figure 1 . The age groups were deliberately selected to be identical in design to those of Brancati and coworkers (7). Figure 2 shows a Kaplan–Meier plot for each age group stratified by the presence or absence of comorbid diseases. Table 1 shows the prevalence of other potential risk factors for subsequent mortality in each age group.
A Kaplan–Meier plot of mortality by PSI grade is shown in Figure 3 . A Cox regression analysis utilizing all components of the PSI score was performed. A summary of the final factors included in the Cox regression model is shown in Table 2 . Treated as a continuous variable, hematocrit was not predictive of mortality (p = 0.14). However, as a categorical variable using the same cutoff of 0.35 as Brancati and coworkers (7), a strong association between hematocrit and subsequent mortality, which remained after adjusting for age, cerebrovascular and cardiovascular disease, and altered mental state, was found (Table 2). Serum glucose on admission was also independently predictive of mortality after correcting for the above-cited factors, but no critical cutoff value was identifiable. Modeling various cutoffs for other physiological variables in the PSI failed to find any significant association.
The development of shock (p = 0.007) or the need for mechanical ventilation (p = 0.009) during the episode of CAP was predictive of higher mortality rates after discharge. However, once age and cardiovascular and cerebrovascular disease were included with these organ failures in a Cox regression analysis, neither shock (p = 0.22) nor mechanical ventilation (p = 0.17) remained a significant factor. To enable comparison with published data on medium-term mortality after intensive care unit admission (4–6), we analyzed survival by APACHE II score (Figure 4) . Survival curves in each APACHE II score class are similar to those previously reported (4–6). A Cox regression analysis found that the point total for all three components of the APACHE II score (A, physiological; B, chronic organ failure; and C, age) (12) were independent predictors of subsequent mortality (p < 0.001 in all cases).
Table 3 shows a comparison of observed mortality rates for all age groups compared with the expected mortality calculated from published U.S. age-, sex-, and race-matched mortality rates (14). As underlying comorbidities are likely to have a significant influence on mortality and we were interested in whether mortality rates were higher than expected in subjects without underlying life-limiting diseases, we defined the subgroup of subjects with no significant comorbidities (including all comorbidities defined in the PSI plus chronic obstructive pulmonary disease and diabetes). An expected mortality over the period of follow-up was again calculated for each subject from published age-, sex-, and race-matched U.S. population data (Table 3) (14).
We have examined the factors predicting mortality in the 2 to 4 years after discharge from hospital with CAP. The major predictors of subsequent mortality were increasing age, the presence of cardiovascular or cerebrovascular disease, and the presence of an altered mental state at the time of presentation to hospital with CAP. Although some correlation between the severity of CAP and subsequent survival was seen, age and comorbidities largely accounted for this association. The primary concern addressed by our study was the issue raised by Brancati and colleagues (7) that an episode of CAP is an adverse prognostic marker for medium- to long-term survival. In all age groups it was not surprising that mortality exceeded that expected from population data because of the high prevalence of life-limiting comorbid diseases. Our primary concern was whether an episode of CAP has adverse implications for longer term survival in the absence of known life-limiting comorbidities. As best demonstrated in Figure 2, no apparent excess medium-term mortality occurred in the 18- to 40-year age group in the absence of comorbid diseases. Although our study is not powered to exclude anything but large mortality differences, this finding is certainly reassuring, particularly given the earlier alarming findings of Brancati and coworkers (7) in this age group. However, there was a trend toward higher mortality in all older age groups with no comorbid illnesses, especially the 41- to 60-year age group. This trend is even more concerning when the fact that the expected mortality rate is based on population data that includes subjects with significant comorbidities is considered. Although the lack of statistical significance may be reassuring, our data do not exclude the possibility that an episode of CAP may be a sentinel event for increased mortality in a subgroup of patients. Although our cohort was both larger and monitored for a longer period of time than that of Brancati and coworkers, we observed a similar overall mortality rate (38 versus 35%) (7). As other studies of medium-term survival after critical illness support our finding of increasing age as a significant predictor of postpneumonia mortality (4–6), we believe the failure of Brancati and coworkers to detect age-related medium-term mortality differences may have been due to an excess of significant comorbidities in younger subjects, relative to our cohort. The finding that both cardiovascular and cerebrovascular diseases increase the subsequent rate of mortality is not surprising. Similarly, poorly controlled diabetes, reflected in the correlation between absolute blood glucose level and mortality, also has strong biological plausibility. What is less clear and requires further comment are the two other significant independent predictors of mortality: an altered mental state at presentation and a hematocrit of less than 35%. Increasing age and underlying organic brain disease are two major risk factors for confusion in any patient hospitalized with sepsis (15). However, the multivariate analysis showed that an altered mental state was a strong predictor of subsequent mortality even after correcting for age and the presence of cerebrovascular disease. Consistent with our findings, a study of elderly patients also identified an altered mental state as a strong predictor of 6-month mortality after discharge from hospital (16). Confusion also appears to be an independent risk factor for death during hospitalization for CAP (17). One explanation is that altered mental state is indicative of some underlying neurodegenerative process not recognized at the time of admission (i.e., Alzheimer's disease or cerebrovascular disease). Confusion at admission may also reflect active and excessive alcohol ingestion (18), with the associated long-term health consequences. Confusion is also more likely to occur in patients with more severe organ disease, especially more severe cerebrovascular disease, which may not be adequately controlled for in our analysis. In any respect, for patients with CAP who present with an altered mental state, physicians should be vigilant for the opportunity to intervene in disease processes that may have been previously unrecognized or the severity underestimated. The association between hematocrit and mortality is much harder to explain. The absence of any correlation when hematocrit is treated as a continuous variable could suggest the association is spurious. However, that the arbitrary cutoff of 35% was chosen due to the identical findings of Brancati and colleagues adds support to the possibility that the association is real (7). Because anemia is a common complication of chronic illness, low hematocrit, like confusion, may function as a marker of more severe chronic illness. Although further study is clearly required, a hematocrit of less than 35% in a patient with CAP, similar to the occurrence of mental confusion, should alert the physician to the possibility of comorbid life-limiting disease processes. The primary limitation of our study is that the actual cause of death in our subjects was not collated. Although potentially available from death certificates, the deficiencies in this approach are well documented (19, 20). As our primary objective was to determine whether an episode of CAP per se had adverse implications for medium-term survival, establishing the exact cause of death was not critical. A further limitation is that the diagnosis of comorbid diseases was based on clinical history, examination, and routine investigations, such as creatinine for renal failure and liver function tests for hepatic failure. Subclinical disease, such as cerebrovascular disease as already discussed above, could clearly have been missed as cranial imaging studies were not routinely performed, nor were echocardiograms or gated heart scans performed to exclude impaired left ventricular function. Although this partially limits the conclusions that can be drawn from our analysis, all the factors incorporated into our statistical models are readily available to physicians treating patients with CAP. Therefore our findings remain highly applicable in the usual clinical context. In summary, we have shown that increasing age, comorbid diseases (especially cardiovascular and cerebrovascular disease), the presence of an altered mental state, a hematocrit of less than 35%, and poor glycemic control are significant and independent predictors of mortality in the subsequent 2 to 3 years after discharge for CAP. Physicians treating patients with CAP should be aware of the predictors of increased mortality, as they may indicate previously unknown or underestimated comorbid diseases. Although the natural course of some of these disease processes may not be alterable, earlier recognition maximizes the potential for interventions to impact on subsequent morbidity and mortality.
Supported by the Methodist Le Bonheur Healthcare Foundation and the Mid-South Pulmonary and Critical Care Research Foundation. Conflict of Interest Statement: G.W.W. has no declared conflict of interest; L.A.K. has no declared conflict of interest; R.G.W. has no declared conflict of interest. Received in original form October 24, 2003; accepted in final form December 22, 2003
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
METHODS
81 (filled triangles) years.
