Modifiable Risk Factors for Nursing Home-Acquired Pneumonia

Abstract

Background. This study sought to identify modifiable risk factors for pneumonia in elderly nursing home residents.

Methods. A cohort of 613 elderly residents (age, >65 years) of 5 nursing homes in the New Haven, Connecticut, area was followed-up prospectively from February 2001 through March 2003. The primary outcome was radiographically documented pneumonia within a 12-month surveillance period. Baseline modifiable risk factors were evaluated for their independent association with pneumonia.

Results. Of 613 elderly nursing home residents, 131 (21%) died, and an additional 112 (18%) developed a radiographically documented case of pneumonia during the 12-month surveillance period. Among the 9 candidate modifiable risk factors that were evaluated individually in Cox proportional hazards models adjusting for covariates (i.e., nursing home facility, age, race, coexisting conditions, and immobility), inadequate oral care (hazard ratio [HR], 1.60; 95% confidence interval [CI], 1.06–2.35; P = .024) and swallowing difficulty (HR, 1.65; 95% CI, 1.04–2.62; P = .033) were associated with pneumonia. When modifiable risk factors were evaluated simultaneously in the same Cox proportional hazards model, inadequate oral care (HR, 1.55; 95% CI, 1.04–2.30; P = .030) and swallowing difficulty (HR, 1.61; 95% CI, 1.02–2.55; P = .043) remained independently associated with pneumonia, adjusting for the same covariates. Calculation of population-based attributable fractions showed that 21% of all cases of pneumonia in our cohort could have been avoided if inadequate oral care and swallowing difficulty were not present.

Conclusions. Two biologically plausible and modifiable risk factors increased the risk of pneumonia in elderly nursing home residents. These results provide a framework for the development and testing of a targeted pneumonia prevention strategy.

Currently, 4.3% of Americans >65 years old reside in nursing homes, and they develop pneumonia at rate of 1 episode per 1000 days of care [1, 2]. This rate is 10-fold greater than the rate of pneumonia in elderly community dwellers [3]. If the incidence of pneumonia in this population remains the same, by the year 2030, there will be an estimated 1.9 million episodes of nursing home—acquired pneumonia annually [4]. Pneumonia is the leading cause of death in elderly nursing home residents, and it results in considerable morbidity, functional decline, and health care expenditures because it is a major cause of transfer to acute care facilities [5–7].

Under the conventional disease model of pneumonia, the invading microorganism is the primary focus of pathogenesis, diagnosis, and treatment. However, in elderly nursing home residents, clinical disease is often multifactorial in origin. For infectious diseases, like pneumonia, the invading microorganism is the final precipitating event superimposed on a set of predisposing risk factors that create a vulnerable host; some risk factors are modifiable and amenable to intervention.

Prevention strategies targeted at modifiable risk factors are needed to reduce the morbidity, mortality, and health care utilization associated with nursing home—acquired pneumonia. In this study, a cohort of elderly nursing home residents was prospectively evaluated to identify and validate modifiable risk factors for pneumonia that are amenable to feasible interventions.

Methods

Participants. Elderly residents (age, >65 years) were eligible for enrollment from February 2001 through March 2002 if they resided in 1 of 5 nursing homes in the New Haven, Connecticut, area for ⩾1 month. Residents who were housed for short-term rehabilitation, whose conditions were deemed terminal by staff, who were fed exclusively by gastrostomy or jejunostomy tube, or who were unwilling to give informed consent were excluded from the study. These criteria were established to study a cohort of residents who were vulnerable to pneumonia, who would have 1 year of surveillance, and who could potentially benefit from a targeted intervention. All participants underwent prospective surveillance for 12 months after enrollment or until there was a documented end point (i.e., pneumonia or a censoring event).

After screening 669 residents, 34 were excluded because they were fed exclusively by gastrostomy or jejunostomy tube, and 22 declined to participate; 613 participants were enrolled in the study. All nursing home facilities and the Human Investigation Committee at Yale University School of Medicine (New Haven) approved the conduct of the study; informed consent was obtained from all participants or from their surrogate decision makers.

Data collection. Information collected at study enrollment included demographic characteristics, degree of mobility, dependence for activities of daily living, and coexisting conditions, as well as the presence of modifiable risk factors for pneumonia. Data were collected from the medical record (including the Minimal Data Set form), as well as from the nursing staff who were involved with the daily care of the resident. Modifiable risk factors were selected on the basis of the available literature, clinical experience, and biological plausibility for pneumonia risk. This set of risk factors included inadequate oral care (defined as lack of dental examination), lack of influenza vaccination within 1 year before enrollment, active smoking, depression, use of sedative medication, use of gastric acid—reducing medication, use of an angiotensin-converting enzyme (ACE) inhibitor, feeding position at an angle of <90° from the horizontal, and difficulty swallowing (defined as cough during swallowing). Reliability testing for these candidate factors was determined by 2 observers in a pilot study and yielded a median κ value of 0.69 (range, 0.58–1.00).

The primary study outcome was the development of a radiographically documented pneumonia, defined as the presence of a compatible infiltrate on chest radiograph (if a previous chest radiograph was available for comparison, the infiltrate must have been new or worsened) and ⩾2 of the following clinical features: new or increased cough, sputum production, shortness of breath, abnormal chest examination findings, pleuritic chest pain, worsening functional status (i.e., decrease in level of consciousness or activities of daily living), fever (temperature, >38°C), or a respiratory rate of >25 breaths/min.

All chest radiographs were interpreted by radiologists involved in the care of the residents at the nursing homes. To test the reliability of the pneumonia outcome, we assessed interrater agreement between 2 of these radiologists in their blinded assessment of radiographic evidence of pneumonia (i.e., new or worsening infiltrate visible on chest radiograph) in a pilot set of 20 residents who were participating in the study. Results of the reliability testing by the 2 radiologists yielded a κ value of 0.57.

Statistical analysis. Proportions or means were used to describe the baseline characteristics of the study population and the incidence of either pneumonia or a censoring event during the 12-month surveillance period. Cox proportional hazards regression was used to examine the relationships between modifiable risk factors and incident pneumonia. Models were adjusted for a priori identified covariates, including study site (5 nursing homes), race (white vs. nonwhite), age (years), immobility, presence of chronic obstructive pulmonary disease, and total number of other comorbidities. Time to event occurrence was defined as days from enrollment to the date of the primary outcome (i.e., first case of radiographically documented pneumonia) or censoring event (i.e., death, discharge from the nursing home facility, or end of follow-up), whichever came first.

In additional Cox modeling, modifiable risk factors were examined simultaneously to determine the subset that demonstrated the strongest association with pneumonia after adjusting for covariates. Backward selection was applied using a retention criterion set at P ⩽ .10. The proportional hazards assumption was evaluated using graphical techniques for each selected modifiable risk factor. The fit of the final model was evaluated using the Akaike Information Criterion and by examining residual and influential observations. Population-based attributable fractions of modifiable risk factors that demonstrated independent association with incident pneumonia were calculated on the basis of the prevalence of the risk factors and the magnitude of the effect estimates (i.e., hazard ratio [HR] for pneumonia) [8]. Because death was a potential competing end point of pneumonia in our study, separate Cox models were examined for each of the baseline modifiable risk factors individually and for the selected risk factors jointly for their association with death after adjusting for the same covariates and ignoring pneumonia. Statistical analyses were conducted using SAS software, version 8.2 (SAS Institute). Bootstrapping methodology was used to establish the internal validity of parameter estimates of the effects of significant modifiable risk factors [9]. One thousand bootstrap samples with replacement were generated and new parameter estimates were obtained from them with use of S-Plus software, version 6.1 (Insightful). Two-sided hypothesis tests were conducted.

Results

Baseline characteristics. Baseline descriptions of the 613 study participants are shown in table 1. The cohort consisted of elderly adults (mean age, 84.7 years), with almost one-half being >86 years of age. Most residents were women (75%); 16% were nonwhite. A large proportion (40%) were immobile (i.e., bed-bound or confined to wheelchairs) and had significant coexisting conditions, including dementia (66%), stroke (25%), congestive heart failure (24%), diabetes (23%), cancer (15%), chronic obstructive lung disease (15%), and kidney disease (11%). Large proportions were dependent in activities of daily living, including bathing (60%), dressing (56%), and feeding (17%); almost one-half had bladder incontinence (49%) or bowel incontinence (45%).

Table 1

Baseline characteristics of study participants residing in 5 nursing homes in the New Haven, Connecticut, area.

Outcomes. Over the 12-month prospective cohort surveillance period, 131 residents (21%) were censored because of death, and 23 residents (4%) were censored because of discharge from the nursing home facility without an episode of pneumonia. A total of 112 residents (18%) developed radiographically documented pneumonia. Of the 112 residents with pneumonia, 46 (41%) of 112 were hospitalized, and 21 (19%) of 112 died; 15 of the 21 residents died within 1 week after pneumonia diagnosis. An additional 123 residents were hospitalized for reasons other than pneumonia.

Association of baseline modifiable risk factors with pneumonia. The prevalence of baseline modifiable risk factors and their association with pneumonia in the 12-month surveillance period are shown in table 2. As shown, the 2 modifiable risk factors significantly associated with pneumonia when evaluated individually in Cox proportional hazards models were inadequate oral care and swallowing difficulty. As shown in table 3, when the modifiable risk factors were examined simultaneously using a Cox proportional hazards model, inadequate oral care (HR, 1.55; 95% CI 1.04–2.30; P = .030) and swallowing difficulty (HR, 1.61; 95% CI, 1.02–2.56; P = .043) remained associated with pneumonia when adjusting for covariates. Bootstrapping of the parameter estimates of these modifiable risk factors suggested bias of very small magnitudes. Specifically, when 1000 samples were constructed with data from study participants and the Cox model was fitted to each sample, the mean of all of the resulting HRs varied from our original estimates by only 1.32% for inadequate oral care and 0.46% for swallowing difficulty. Cox modeling using death as the end point revealed a significant association only with cough during swallowing (HR for cough during swallowing, 1.62 [P = .015]; HR for inadequate oral care, 1.24 [P = .230]).

Table 2

Proportional hazards models for the association of individual modifiable risk factors with incident pneumonia during a 12-month surveillance period (n = 613).

Table 3

Association of 2 selected modifiable risk factors with incident pneumonia analyzed simultaneously in proportional hazards model.

Estimation of attributable fractions. From the prevalence data and adjusted HRs in the final Cox proportional hazards model, calculations were made of the fractions of pneumonia cases that were attributable to each of the 2 independent modifiable risk factors. As shown in table 3, the fraction of cases of pneumonia in the entire cohort that could have been prevented if inadequate dental care had not occurred was 0.16 (or 16%); the fraction that could have been prevented if swallowing difficulty had not occurred was 0.05 (or 5%). Among the subset of the cohort who possessed ⩾1 of the 2 risk factors at baseline, 35% of pneumonia cases could have been prevented if inadequate dental care had not occurred, and 38% of pneumonia cases could have been prevented if swallowing difficulty had not occurred.

Discussion

In this study of 613 elderly nursing home residents, 112 (18%) developed a radiographically documented pneumonia during a 12-month surveillance period. Statistically significant independent association with pneumonia was shown for 2 modifiable risk factors: inadequate oral care and swallowing difficulty. Internal validation using bootstrapping demonstrated minimal bias of the effect estimates of each modifiable risk factor. Attributable fraction measurements revealed that up to 21% of pneumonia cases in the entire cohort could have been prevented if both modifiable risk factors had not occurred.

Nursing home—acquired pneumonia was described in 1978, when the etiologic bacteria (i.e., Klebsiella pneumoniae and Staphylococcus aureus) were recognized as distinct from those causing community-acquired pneumonia [10, 11]. In a 1981 multicenter survey of infections among elderly nursing home residents [12], it was first suggested that pneumonia was associated with modifiable risk factors, including use of sedative medication, overuse of antibiotics, use of suboptimal feeding techniques, and inconsistent immunization practices. Subsequent investigations involving elderly nursing home residents have identified potential risk factors for pneumonia, including tube feeding [13, 14, 16], depression [13], reduced level of consciousness [13, 14], malnutrition and weight loss [14], mobility limitation [15], smoking [16], lack of vaccinations [17], inadequate dental care [16, 18], sedative medications [16, 18], and difficulty swallowing [17–19].

Compared with previous work, our study identified 2 similar areas of modifiable risk for pneumonia (i.e., inadequate oral care and swallowing difficulty), but it did not find associations with other modifiable risk factors for pneumonia. This was potentially the result of our explicit definition of risk factors and outcome or of the lack of an independent association when appropriate multivariable models were used. The methodologically unique strengths of our study included (1) cohort assembly from 5 community nursing homes (i.e., there was no restriction to academically affiliated facilities), which resulted in a study population that was representative of the demographic characteristics and coexisting illnesses of contemporary elderly nursing home residents; (2) comprehensive assessment of modifiable risk factors (i.e., 9 candidate modifiable factors were assessed in multivariable modeling); (3) use of a rigorous pneumonia outcome definition (i.e., including radiographic documentation); (4) interobserver reliability testing of both modifiable risk factor assessment and pneumonia outcome detection; (5) use of an analytic model that included information on subjects who died or who were lost to follow-up before pneumonia was observed; (6) use of bootstrapping as a measure of validation; and (7) defining the attributable fraction of cases of pneumonia that could be prevented by modifying each risk factor.

Our comprehensive examination of modifiable risk factors allowed us to identify 2 factors showing independent association with pneumonia that are both biologically plausible. Lack of dental examination is common in nursing home populations (49% in our cohort), and it is a plausible surrogate for inadequate oral care. Colonization with periodontal pathogens (e.g., Prophyromonous gingivalis) can degrade fibronectin on oral mucosal epithelium and enable colonization with potential respiratory pathogens (e.g., gram-negative bacilli and S. aureus) [20]. Previous studies support the association of inadequate oral care with pneumonia in both hospitalized patients and elderly nursing home residents [20, 21], with additional evidence suggesting that oral care interventions may reduce pneumonia risk [22]. Impaired swallowing and cough reflexes have been identified in older adults who develop aspiration pneumonia [23, 24]. Animal models and human studies have suggested that swallowing and cough reflexes are governed by a plexus of nerves containing the hormone called substance P [25, 26]. Experimental reduction of substance P using dopamine antagonists impairs swallowing [27], and patients with impaired dopamine production associated with basal ganglia infarcts have delayed swallowing reflexes and a higher risk of pneumonia [28]. Interventions that increase substance P, including nonpharmacologic methods (e.g., capsaicin and gum stimulation by aggressive oral care) and pharmacologic means (e.g., levodopa and ACE inhibitors), have suggested a therapeutic benefit in humans [29–31]. ACE inhibitor use was reported to be associated with a reduced risk of pneumonia in patients with stroke, compared with use of other antihypertensive drugs [32]. Although not statistically significant in our cohort when controlling for other modifiable risks and baseline covariates, the relative risk of 0.71 for ACE inhibitors corroborates this finding.

Despite the advantages of our large prospective cohort evaluation and the rigorous outcome ascertainment, there were limitations to our study. First, lack of dental examination was the only surrogate we used to reflect inadequate oral care. Clinical measures, including presence of teeth and denture use, were analyzed in our study, but they did not show an association with pneumonia in individual Cox modeling, nor did they show any correlation with lack of dental examination. Other, more specific measures (e.g., the frequency of gum or tooth brushing, degree of dental caries, number of teeth, degree of periodontal inflammation, degree of bone loss, and pocket depth) may have shown stronger association with pneumonia. Future cohorts should be evaluated for the impact of these clinical oral hygiene variables on pneumonia risk, although the reliability and feasibility of ascertainment would require confirmation. Second, although cough during swallowing represents the best bedside surrogate to predict aspiration documented by videofluoroscopy [33], individuals may aspirate oropharyngeal contents silently, and such individuals may have been unrecognized in our cohort [34]. If true, this would have increased the risk of pneumonia in our comparative group of residents with no identifiable risk factors and suggests that our observed HRs were conservative underestimations of the actual risk. Third, our assessment of modifiable risk factors was made only at baseline. Although a pilot evaluation revealed no major changes in risk factors 6 months after baseline, a more comprehensive assessment of a larger cohort may demonstrate dynamic changes in risk after baseline determination; this may further explain the 14% risk of pneumonia for residents with no identifiable risk factor at baseline. Finally, although bootstrapping techniques provided a measure of internal validation of our results, external validation in other nursing home cohorts would be valuable.

These limitations notwithstanding, our results identified and validated 2 biologically plausible and modifiable risk factors for pneumonia in a large prospective cohort of elderly nursing home residents. Although the impact of each modifiable risk factor on the development of pneumonia was modest, almost one-half of the residents (285 [47%] of 613) had ⩾1 of the 2 risk factors at baseline. This finding suggests that nursing home—acquired pneumonia, similar to other common morbid health problems in older persons, can be conceptualized as a geriatric syndrome that results from the effect of >1 predisposing factor within the individual. Therefore, it may be possible to reduce this risk by modifying one or both of these factors through a targeted intervention strategy. As demonstrated in the calculation of attributable fractions, up to 21% of all cases of pneumonia observed in the total cohort could have been prevented if the 2 major risk factors were avoided; this represented more than one-third of the cases of pneumonia among the subset of residents with ⩾1 of the 2 risk factors at baseline. The finding that 1 of the risk factors (i.e., cough during swallowing) showed a significant positive association with death suggests that successful modification of these risk factors will not increase—and may decrease—risk of death. Similar approaches aimed at determining modifiable risk factors for other multifactorial geriatric syndromes (e.g., falls in the community and delirium during hospitalization) have resulted in clinically successful and cost-effective targeted preventive strategies [35–39].

Given the potential direct impact of aggressive oral care (e.g., gum and tooth brushing) on oral hygiene and its indirect impact on the swallowing reflex (through an increase in the production of substance P), it is plausible that a single, feasible intervention may reduce nursing home—acquired pneumonia. As a result of our study, future investigations should evaluate candidate oral care intervention options (e.g., daily gum and tooth brushing) for their feasibility and staff adherence within the nursing home environment, their effectiveness in improving oral hygiene and swallowing, and their ultimate ability to prevent pneumonia in this vulnerable population.

Acknowledgments

Financial support. National Institutes of Health/National Institute on Aging Claude D. Pepper Older Americans Independence Center (grant P30-AG21342) and the John A. Hartford Foundation.

Potential conflicts of interest. All authors: no conflicts.