Legionnaires' Disease and Associated Comorbid Conditions as Causes of Death in the U.S., 2000–2010

Abstract

Objective

Recent U.S. outbreaks of Legionnaires' disease (LD) underscore the virulent nature of this infectious pneumonia. To date, only a paucity of literature has described the mortality burden of LD. This study updates LD mortality using U.S. multiple-cause-of-death data from 2000–2010.

Methods

We calculated crude and age-adjusted rates for LD mortality for age, sex, race, state, Census region, and year. We conducted Poisson regression to assess seasonal and temporal trends. We generated matched odds ratios (MORs) to describe the association between LD-related deaths and other comorbid conditions listed on the death certificates.

Results

We identified a total of 1,171 LD-related deaths during 2000–2010. The age-adjusted mortality rate remained relatively static from 2000 (0.038 per 100,000 population, 95% confidence interval [CI] 0.031, 0.046) to 2010 (0.040 per 100,000 population, 95% CI 0.033, 0.047). The absolute number increased from 107 to 135 deaths during this period, with adults ≥45 years of age having the highest caseload. Overall, LD mortality rates were 2.2 times higher in men than in women. White people accounted for nearly 83.3% of all LD-related deaths, but the age-adjusted mortality rates for black and white people were similar. Comorbid conditions such as leukemia (MOR=4.8, 95% CI 3.5, 6.6) and rheumatoid arthritis (MOR=5.6, 95% CI 3.3, 9.4) were associated with LD diagnosis on death certificates.

Conclusion

Comorbid conditions that could lead to an immunocompromised state were associated with fatal LD on U.S. death certificates. Characterization of LD mortality burden and related comorbidities has practice implications for clinical medicine and public health surveillance.

Legionnaires' disease (LD) is a severe form of pneumonia that can become fatal in vulnerable individuals.^1–5 The condition is caused by the gram-negative bacterium Legionella pneumophila, a naturally occurring organism found in water.^2,3 The organism is primarily transmitted to humans through inhalation of contaminated aerosolized water droplets. Typical sources of outbreaks include, but are not limited to, hot tubs, hot water tanks, large plumbing systems, decorative fountains, and cooling towers. The pneumophila species is generally not found in car or window air-conditioners.^1–4 To date, no person-to-person transmission has been documented.^1–4,6,7

Risk factors that increase susceptibility to LD include advanced age, smoking, chronic lung disease, and having a weak or suppressed immune system.^1–4,6,7 Among individuals with a weakened immunity, those with cancer, diabetes, or kidney failure are particularly at risk. Although human immunodeficiency virus (HIV) infection has been linked to more severe infections, LD is generally not more common among people with HIV/acquired immunodeficiency syndrome (AIDS).^1,8,9 More than 20% of cases in the United States reported to the Centers for Disease Control and Prevention (CDC) each year are travel-related.³ Nosocomial transmissions of LD can be problematic, as at-risk individuals are plentiful and plumbing in hospital facilities is often old.^1,4 Because in-house diagnostic testing is readily available, detection of LD is more likely in the hospital setting.⁴ Currently, CDC recommends urinary antigen assay and a culture of respiratory secretions on selective media as the diagnostic tests of choice for confirming Legionella infection.³

Annually, CDC estimates that as many as 18,000 individuals are hospitalized with LD in the U.S.³ However, this number is considered an underestimate, as the disease is often underreported due in part to its diagnostic complexity and its resemblance to other pneumonias with similar symptom presentations.^3,7 Symptoms of LD include cough, shortness of breath, fever, muscle aches, and headaches; they often start shortly after exposure to the bacteria.^1–4,6

Recent LD outbreaks in the Veterans Affairs (VA) Medical Center in Pittsburgh, Pennsylvania, and in a retirement community in Ohio affirm the virulent nature of this infection and underscore the importance of the causative infectious pathogen to clinical practice and public health surveillance.^10,11 Based on hospitalizations and treatment data, the annual costs for LD have been estimated at nearly $684 million.¹² Few studies have recently examined LD mortality and its associated comorbid conditions. This study contributes to the closing of this gap in the literature by analyzing the national multiple-cause-of-death (MCD) data to describe the current mortality profile of LD in the U.S., for the period 2000–2010.

METHODS

We performed trend and matched case-control analyses using MCD data for the years 2000–2010 obtained from the National Vital Statistics System. The National Center for Health Statistics (NCHS), which manages this system, collects data on reported deaths filed in all 50 U.S. states, including the District of Columbia.^13,14 Death occurrences are generally reported to NCHS using the U.S. Standard Certificate of Death form (i.e., death certificate). The attending physician or coroner usually completes this form, documenting decedent demographics and information about the underlying and associated causes of death as determined by the reporting party.^13,14

Causes of death from 1999 to 2010 were classified in accordance with the International Classification of Diseases, 10th Revision (ICD-10).¹⁵ If more than one cause was listed on a death certificate, the sequence of conditions or causes on the form, provisions of the ICD, and relevant selection rules were applied to arrive at the underlying cause.^13,14 Studies using MCD data to detect associations and causality based on conditions listed on death certificates have been previously described by Redelings et al.¹⁶

Study population and data analysis

We identified underlying and associated causes of LD using the ICD-10 code A48.1; these cases were subsequently included in the analysis. We calculated LD mortality rates per 100,000 population using the bridged-race population estimates provided by NCHS and population data from the U.S. Census Bureau.^17,18 We calculated crude rates and 95% confidence intervals (CIs) for age, sex, race/ethnicity, state of residence, and Census region. We estimated age-adjusted rates, ratios, and 95% CIs using the 2000 U.S. standard population for sex and race/ethnicity. Where appropriate, we described and analyzed seasonal and temporal trends using Poisson regression procedures.¹⁹

Matched case-control analysis

To assess comorbid conditions as possible factors contributing to fatal LD, 10 controls (i.e., non-LD-related deaths) were randomly selected for each LD-related death from the other death records in the MCD dataset and matched on age at death, sex, and race/ethnicity. Comorbid conditions were selected for inclusion in the matched analysis if they met one or more of the following criteria: (1) the condition was listed on ≥20 death certificates in which LD was listed as an underlying or associated cause of death, (2) the condition has been previously found to be associated with the contraction of LD (based on a literature review), or (3) the condition likely contributed to the decedent being immunocompromised. Septicemia, adult respiratory distress syndrome, respiratory failure, and shock were considered possible sequelae of severe LD and, therefore, were excluded from the analysis. For each comorbid condition, matched odds ratios (MORs) and accompanying 95% CIs were generated and reported using previously established methods.¹⁶ We conducted all MCD analyses using SAS^® version 9.3.²⁰

RESULTS

We identified 1,171 deaths from LD during the study period 2000–2010. LD was listed as an underlying cause of death for 922 (78.7%) of the total decedents and as the associated cause for the remainder of decedents (21.3%) (data not shown).

The annual age-adjusted mortality rate for LD was relatively static during the sampled period, increasing only slightly from 0.038 per 100,000 population (95% CI 0.031, 0.046) in 2000 to 0.040 per 100,000 population (95% CI 0.033, 0.047) in 2010. The absolute number of deaths increased from 107 in 2000 to 135 in 2010 (Figure 1). The overall age-adjusted mortality rate, averaged over the entire sampled period, was 0.035 per 100,000 population (95% CI 0.033, 0.037). Time-trend analysis showed an average annual percentage change in mortality rates of –1.22% (95% CI –0.60, –3.08) (data not shown). A seasonal trend analysis revealed that nearly half (47.3%) of all deaths during the sampled period occurred from July to October (Figure 2).

Figure 1.

Number of deaths from Legionnaires' disease and age-adjusted mortality rates per 100,000 population, by year: U.S. multiple-cause-of-death data, 2000–2010 (n=1,171)

Figure 2.

Frequency distribution of deaths from Legionnaires' disease, by month: U.S. multiple-cause-of-death data, 2000–2010 (n=1,171)

The mean age at death for all LD-related deaths was 66.1 years (data not shown). The age-specific mortality rates increased consistently with each increasing age class. Those aged ≥85 years had the highest age-specific mortality rate (n=123; 0.226 per 100,000 population; 95% CI 0.186, 0.266), while those aged 75–84 years had the highest absolute number of deaths (24.2% of the total sample). Nearly two-thirds (63.4%) of the LD cases were men. In comparison analysis, men had a higher rate of LD mortality than women (age-adjusted rate ratio [RR] = 2.2, 95% CI 2.0, 2.4). Although white people accounted for nearly 83.3% of LD-related deaths, age-adjusted mortality rates between black and white people were similar (RR=1.0, 95% CI 1.0, 1.1). Corresponding age-adjusted mortality rates for Asian/Pacific Islanders (RR=0.4, 95% CI 0.4, 0.5), Native Americans (RR=0.4, 95% CI 0.3, 0.5), and Hispanic people (RR=0.4, 95% CI 0.4, 0.5) were substantially lower than for white people (Table 1).

Table 1.

Age-adjusted Legionnaires' disease mortality rates per 100,000 population and mortality rate ratios by region, sex, race/ethnicity, and age group: U.S. multiple-cause-of-death data, 2000–2010 (n=1,171)

^aMortality rates for age groups are age-specific rates, not age-adjusted rates.

CI = confidence interval

Ref. = reference group

NA = not applicable

Geographically, the Census region with the highest age-adjusted mortality rate was the Northeast region (n=311; 0.047 per 100,000 population; 95% CI 0.042, 0.052). The region with the highest absolute number of deaths, however, was the South (n=378; 0.031 per 100,000 population; 95% CI 0.028, 0.034) (Table 1). The state with the highest age-adjusted mortality rate was Delaware (n=12; 0.127 per 100,000 population; 95% CI 0.055, 0.199) (Figure 3), but the state with the highest number of LD-related deaths was Ohio (n=115) (data not shown). The only state in the West region that fell into the fourth quartile of the age-adjusted rates with more than five deaths was Utah (Figure 3); however, the number of deaths was still small (n=16; 0.074 per 100,000 population; 95% CI 0.037, 0.111) (data not shown).

Figure 3.

Age-adjusted rates per 100,000 population of Legionnaires' disease mortality, by state: U.S. multiple-cause-of-death data, 2000–2010 (n=1,171)

In the matched case-control analysis, a number of comorbid conditions were found to be associated with LD-related deaths after matching on age, sex, and race/ethnicity. All cases of LD were successfully matched to 10 controls, resulting in a selection of 11,710 total controls. Using previously described selection criteria, 15 comorbid conditions were selected for analysis because they appeared on ≥20 death certificates, four conditions were selected because the literature showed a positive correlation between their presence and LD severity, and nine conditions were selected because disease progression and effects from treatments could contribute to the decedent becoming immunocompromised. At least 10 conditions met more than one of the selection criteria.

Although all cancer types (MOR=0.5, 95% CI 0.4, 0.5) had lower odds of appearing alongside LD on the death certificates, subgroup analysis revealed that this more generalized observation about cancer was not the case for leukemia (MOR=4.8, 95% CI 3.5, 6.6), multiple myeloma (MOR=2.5, 95% CI 1.4, 4.3), and non-Hodgkin's lymphoma (MOR=2.6, 95% CI 1.7, 4.0). Other comorbid conditions that were positively associated with LD-related deaths included connective tissue disorders (MOR=6.1, 95% CI 4.0, 9.3), specifically rheumatoid arthritis (MOR=5.6, 95% CI 3.3, 9.4), drug and alcohol abuse (MOR=2.8, 95% CI 2.2, 3.5), and all types of renal failure (MOR=3.2, 95% CI 2.7, 3.7) (Table 2).

Table 2.

Comorbid conditions associated with Legionnaires' disease mortality: U.S. multiple-cause-of-death data, 2000–2010

^aAcute and other renal failure can occur as a result of Legionnaires' disease.

ICD-10 5 International Classification of Diseases, 10th Revision

CI 5 confidence interval

DISCUSSION

These study findings suggest that while annual age-adjusted mortality rates for LD in the United States have remained relatively static, the absolute number of cases increased slightly from 2000 to 2010. Decedent characteristics that suggest strong risk for severe disease and death were generally consistent with those described in the literature.^1,3,4,6 For example, Fields et al. identified males as having a higher risk for LD than females;¹ this finding is similar to our finding that men had higher LD mortality rates than women. Age-specific mortality rates likewise increased with each increasing age category, with adults aged ≥45 years representing the largest proportion of the cohort; this finding was not dissimilar to what has been described previously.^1,3 This trend could increase as the U.S. population ages.^21,22

Collier et al. estimated health-care expenditures for LD to be approximately $684 million annually.¹² This estimate, however, did not take into consideration lost productivity due to work absenteeism and losses to society. In a supplemental analysis of the MCD data using the human capital approach,²³ the cumulative productivity losses associated with premature death from LD were projected to be more than $463 million during 2000–2010.

The higher number of deaths observed from July to October was consistent with the increased incidence of LD in summer and fall months seen in prior studies.^7,24 Studies have found that warm, humid, and wet weather conditions can contribute to increased growth of Legionella bacteria; thus, in places such as The Netherlands and certain parts of the U.S., LD incidence was generally higher during summer and fall.^25–28 Not coincidentally, the use of cooling towers was likely higher during these seasons as well.

Although there were notable differences in mortality trends by geography, these increased rates and numbers in certain parts of the country likely reflected clustering of the demographics, seasonality, and other regional risk factors rather than the emergence of more virulent Legionella strains. The Eastern and Midwestern states, for example, have historically had a higher incidence of LD;^24,29,30 this higher incidence may reflect seasonal as well as housing factors, such as humid conditions during the summer months in combination with older plumbing systems in older buildings.

Previous research has implicated a variety of chronic conditions that can impair the immune response as strong risk factors for developing more severe disease and death, independent of age, sex, and race/ethnicity.^1,7,8,24 Our findings affirm these observations in the literature, demonstrating associations between -LD-related deaths and reports of hematological cancers, connective tissue disorders, and drug and alcohol abuse on the death certificates. The latter association, however, should be interpreted with caution, as a mediating factor that can affect the pathway to death is smoking status. Prior studies have shown that drug and alcohol abuse is positively correlated with cigarette smoking,^31–34 and cigarette smoking, in turn, is an established risk factor of LD.^3,4,6,7

Treatment of certain diseases can increase susceptibility to infection. For instance, in rheumatoid arthritis, treatments such as glucocorticoids and tumor necrosis factor inhibitors have been known to increase patient risk for bacterial infections.^35,36 Standard treatments for other connective tissue disorders could carry similar risks.

Consistent with the literature, fatal LD was not more frequent among individuals with HIV/AIDS in the study sample. Although further research can provide additional insights on this finding, medical treatment advances in HIV/AIDS may have affected LD's virulence in the presence of HIV. For example, the widespread use of triple antiretroviral therapy among HIV patients (in the mid-1990s) could have played an instrumental role in reducing the number of individuals susceptible to Legionella infection or to severe manifestations of this disease.⁹

Limitations

Although there are several advantages to using population-level data to describe disease burden (e.g., larger sample size and greater statistical power), there were a number of limitations related to the use of death records.¹⁶ First, due to the study design, it was difficult to show causality because temporality cannot be established among the various factors that could have contributed to death. Second, user errors while completing the death certificates may have resulted in misclassification bias.^37,38 Completeness for each death certificate form typically depends on the physician or coroner who files the records (i.e., the reporting party was responsible for determining the sequences of events and conditions that may have led to death). Third, information on certain risk factors, patients' medical history, the clinical course of the disease (including virulence of the organism), and social characteristics were important data that were not listed or described on the death certificates. Without such information, the MCD analyses were limited in their ability to adequately adjust for confounding. Better technology/protocols and vigilance for disease detection in subsequent years, for example, could have accounted for some of the observed increases in absolute LD numbers during the sampled period. Finally, the population data used and the mortality estimates generated by this analysis may lack a certain degree of precision due to numerous uncertainties. Among them were the high frequency in which LD is often undetected or underdiagnosed in real-world clinical settings and the paucity of definitive guidance on whether and how bias from multiple comparisons could be addressed in studies other than confirmatory clinical trials.³⁹ Among researchers, there are different viewpoints regarding whether and how to adjust for the latter.^40,41

CONCLUSION

We found that having comorbid conditions that could lead to an immunocompromised state was associated with LD-related death on the death certificates. Recent LD outbreaks observed among patients at the Pittsburgh VA and residents of the retirement community in Ohio^10,11 (which collectively took more than 10 lives) affirms the importance of early diagnosis, treatment, and prevention of LD in the hospital and community settings. Based on MCD data, LD-related mortality was not trivial during 2000–2010. These study findings have practical implications for health services planning, clinical medicine, and public health surveillance. Legionella infection is a preventable and treatable disease that remains a formidable public health problem in the U.S.