INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Despite the fact that Pneumocystis carinii pneumonia (PCP) remains the leading acquired immune deficiency syndrome (AIDS)-defining opportunistic infection in the United States (1), little is known about the epidemiology of P. carinii. Traditionally, scientists have believed that PCP results from reactivation of latent infection (2). However, several lines of evidence suggest that active acquisition of P. carinii from environmental exposure or person-to-person transmission may play a role in the development of the disease. Multiple case series have documented outbreaks of PCP among immunocompromised patients (6). Animal studies have also shown that transmission from infected to noninfected rodents can occur via an airborne route, and immunosuppressed animals housed in sterile environments do not develop PCP (9, 10). Further, there have been several recent reports identifying P. carinii DNA in both outdoor and hospital air samples, as well as data suggesting a link between soil exposure and development of PCP (11). These studies raise interesting questions about possible routes of P. carinii transmission, but the answers to these questions have been difficult to find.

Geographic analysis is one technique that can address questions about environmental or human sources of disease. Diseases such as tuberculosis and toxoplasmosis have been routinely studied using geographic analyses (17), but these techniques have not been applied to PCP. The objective of the present study was to examine the influence of geographic location as a risk factor for PCP. We performed a retrospective case-control study of human immunodeficiency virus (HIV)- infected persons evaluated for PCP at San Francisco General Hospital (SFGH) between January 1996 and June 1999.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

PCP Subjects

PCP subjects were HIV-infected adult patients diagnosed with microscopically confirmed PCP at San Francisco General Hospital from January 1996 through June 1999. PCP subjects were identified through the Division of Pulmonary and Critical Care Medicine's computerized log of PCP evaluations. This list was cross-referenced with a list generated from a computerized search of SFGH medical records to ensure the completeness of our PCP subject list. Subsequent P. carinii microscopic diagnoses that occurred within 90 d of a previous PCP diagnosis were excluded from analysis due to our inability to determine if these episodes represented new or recurrent disease. Subjects who were homeless or who lived outside of San Francisco were also excluded from analysis.

Control Subjects

Control subjects were HIV-infected adult patients who had a clinical presentation suggestive of PCP during the same time period as the PCP subjects, but who had a bronchoscopy with bronchoalveolar lavage (BAL) examination that was negative for P. carinii. These subjects were identified through the Division of Pulmonary and Critical Care Medicine's computerized log of PCP evaluations.

These patients were felt to be appropriate controls because they were HIV infected, suspected to have PCP, and underwent the same evaluation as the PCP subjects. At SFGH and its affiliated community health clinics, any patient whose clinical and radiographic presentation suggests PCP undergoes sputum induction (SI) followed by bronchoscopy with BAL if the SI examination is negative for P. carinii. A patient whose BAL fluid examination is negative for P. carinii is felt unlikely to have PCP, and discontinuation of specific PCP treatment is recommended. A prior analysis has shown that the use of SI followed by BAL diagnoses 99.8% of all cases of PCP and that a negative BAL fluid examination for P. carinii virtually rules out the diagnosis of PCP (20). Because the sensitivity of SI for P. carinii is insufficient to rule out the diagnosis of PCP at our institution, no subjects who had a negative SI examination for P. carinii, but who failed to undergo bronchoscopy were included. Control subjects who were homeless or lived outside of San Francisco were also excluded.

Data Collection

Medical chart review of all PCP and control subjects was performed using a standardized data abstraction form. To ensure uniformity of chart abstraction, two study investigators independently reviewed approximately 40% of the charts. Abstracted data included subjects' age, sex, race, ethnicity, and HIV risk group. The subjects' address at time of PCP evaluation was recorded. Finally, details on clinical factors that influence risk of PCP including use of P. carinii prophylaxis, use of antiretroviral therapy, CD4 cell count, and HIV RNA level were collected. Subjects were classified as using P. carinii prophylaxis if they reported taking at least 50% of the doses of a prescribed acceptable anti-Pneumocystis regimen. Patients were considered to be using antiretroviral therapy if they reported consistent use of antiretroviral medication over the preceding 30 d. Given the changing standards for antiretroviral therapy over the course of the study, the use of any antiretroviral medication, alone or in combination, was considered antiretroviral therapy. The most recent laboratory CD4 cell count and HIV RNA that preceded the PCP evaluation were recorded. For persons using antiretroviral therapy, these values were invariably within 90 d preceding evaluation. If no value was available within a year preceding evaluation, the closest value within 30 d after the evaluation was recorded. If there was no laboratory CD4 count or HIV RNA data measured within these times, no data were recorded. Information on the median annual income of persons residing in specific zip codes was obtained from data from the 1990 United States Census (21).

Statistical Analysis

Data were entered twice to ensure accuracy of data entry. Analysis was performed using the Statistical Analysis System (SAS) (22). Univariate analyses were performed using the Fisher's exact and chi-square tests. For the Fisher's exact test, p values were based on the two-tailed test results. Odds ratios and 95% confidence intervals were calculated to assess the univariate risk of a subject being diagnosed with PCP. For analyses involving subject address, the patient's zip code was used. Only zip codes that had at least 20 total evaluations (5% of total study population) were analyzed independently in order to minimize error. These zip codes represent the areas of the city that include approximately 70% of the AIDS population (San Francisco Department of Public Health AIDS Registry). Variables that had a p value  0.10 in the univariate analysis were included in stepwise forward and backward multivariate logistic regression analyses. For all analyses, a p value < 0.05 was used to define significance.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

PCP Subjects

During the 3¹/₂-yr study period, 278 subjects were diagnosed with PCP, and 246 had a San Francisco address at the time of diagnosis. Thirty-two subjects who were diagnosed with PCP were excluded from analysis due to homelessness (n = 26, 9.4%) or residence outside of San Francisco (n = 6, 2.2%). The mean subject age was 38.8 yr (Table 1). The majority of the subjects were male (94%), white (55%), and were men who have sex with men (MSM) (68%). P. carinii prophylaxis (24%) and antiretroviral therapy (17%) use were reported in less than one-quarter of the patients. The mean CD4 cell count was 53 cells/µl; two-thirds of the subjects had a CD4 cell count  50 cells/µl and greater than 95% had a CD4 cell count  200 cells/µl. The mean log HIV RNA level was 11.3 copies/ml.

Six San Francisco zip codes had at least 20 PCP evaluations performed over the course of the study (Table 1). These zip codes had an average of 29.3 cases of PCP (range 19-51 cases) and, combined, accounted for 176 of the 246 (72%) PCP cases for which there were San Francisco addresses at the time of diagnosis. The remaining 19 San Francisco zip codes had an average of 3.7 PCP cases (range 0-12) diagnosed over the course of the study.

Control Subjects

During the same period, 160 subjects had a BAL fluid examination that was negative for P. carinii, and 147 had a San Francisco address at the time of evaluation. Thirteen control subjects were excluded either because they lived outside of San Francisco (n = 11, 6.9%) or were homeless (n = 2, 1.3%). The mean age of the patients was 40.4 yr (Table 1). Similar to the PCP subjects, the majority of the control subjects were male (84%), MSM (56%), and white (50%). P. carinii prophylaxis (70%) and antiretroviral therapy (41%) use were reported in a higher proportion of control subjects than PCP subjects. The mean CD4 cell count was 111 cells/µl; 44% of the subjects had a CD4 cell count  50 cells/µl and 84% had a CD4 cell count  200 cells/µl. The mean log HIV RNA level was 10.5 copies/ml.

The six San Francisco zip codes in which there were at least 20 PCP evaluations performed over the course of the study had an average of 17.3 control subjects without PCP (range 9- 27 controls). Combined, these zip codes accounted for 104 of the 147 (71%) control subjects for which there were San Francisco addresses at the time of diagnosis. The remaining 19 San Francisco zip codes had an average of 1.5 control subjects without PCP (range 0-12) evaluated over the course of the study.

Thus, greater than 70% of the subjects with and without PCP resided in the six zip codes that accounted for approximately 70% of the AIDS population in San Francisco. In addition, the proportion of PCP subjects diagnosed (72%) and the proportion of control subjects without PCP evaluated (71%) in the six zip codes selected for further analysis were virtually identical (p > 0.5).

Influence of Zip Code on Probability of PCP

Univariate analysis demonstrated that there were several variables associated with an increased probability of PCP (Table 2). The demographic characteristics of male sex, a history of MSM, and no history of injection drug use were all associated with an increased probability of PCP. In addition, the clinical characteristics of lack of P. carinii prophylaxis or antiretroviral therapy, a CD4 cell count  50 cells/µl, and a higher mean log HIV RNA level were also all strongly associated with an increased probability of PCP.

Of the six zip codes, the only one that had a significant association with PCP probability was zip code 94103 (Table 2). Interestingly, subjects residing in this zip code were less likely to have PCP than those residing in other zip code areas (odds ratio 0.5, 95% confidence interval 0.3-0.8, p = 0.01) (Figure 1).

View larger version (53K): [in this window] [in a new window]   Figure 1.   Map of San Francisco showing the six zip codes with greater than 20 PCP evaluations performed from January 1996 through June 1999.

Multivariate Analysis

Multivariate analysis revealed that residence in zip code 94103 remained independently associated with a lower probability of PCP (Table 3). Subjects residing in this zip code were one-fifth as likely to have PCP as those residing in other zip code areas when the analysis controlled for factors including race, HIV risk category, use of PCP prophylaxis or antiretroviral therapy, CD4 cell count, and HIV RNA level (adjusted odds ratio = 0.2, 95% confidence interval = 0.1- 0.6, p < 0.01). We explored several different multivariate models; in all of them, residence in this zip code remained an independent predictor of a lower likelihood of PCP (range of adjusted odds ratios 0.2-0.4; data not shown).

The possibility that a subject's zip code was merely a marker for socioeconomic status and that lower socioeconomic status rather than geographic area was actually the risk factor for PCP was explored. Our assumption was that patients with advanced AIDS and PCP might have been forced to move from higher to lower income housing as their HIV disease became more severe. Under this scenario, one would expect that subjects residing in zip code 94103 would have a higher income than subjects living in the other zip codes; however, examination of median annual household incomes for the six zip codes revealed that zip code 94103 was among the poorest areas of the city. The median annual household income in the zip code was $18,411 compared with an average of $29,094 for the other five zip codes (21).

To examine if a higher incidence of respiratory infections other than PCP could explain the results in the 94103 zip code, we recorded all diagnoses made in the control subjects. There were no significant differences among the diagnoses made in the 94103 control subjects and the control subjects from all other zip codes. Bacterial pneumonia accounted for 22% of the diagnoses in 94103 and 26% in the other zip codes (p = 0.9). Tuberculosis (TB) was uncommon in all areas (0% of 94103 diagnoses, 3% of all others, p = 1.0). This low rate of TB was not due to a low rate in San Francisco's HIV-infected population, but rather because patients with clinical and radiographic presentations consistent with TB were not evaluated for PCP and therefore not included in the study population. Miscellaneous respiratory conditions such as fungal pneumonia, atypical mycobacteria, Kaposi's sarcoma, and congestive heart failure (CHF) accounted for the majority of all remaining diagnoses (48% of 94103 diagnoses, 42% of other, p = 0.62).

To determine if the control subjects in zip code 94103 differed in any systematic way from the PCP subjects in that zip code, we compared the demographic characteristics of these two groups (Table 4). Age, sex, race, and HIV risk group did not differ between the two groups (p > 0.4 for all comparisons).

In our multivariate analysis, the other independent predictors of PCP were no P. carinii prophylaxis use and a CD4 cell count  50 cells/µl. Subjects not using P. carinii prophylaxis were nine times more likely to have PCP than those using prophylaxis (adjusted odds ratio = 9.0, 95% confidence interval = 4.7-17.5, p < 0.01) whereas those with a CD4 cell count  50 cells/µl were more than four times as likely to have PCP compared with those with a CD4 count > 50 cells/µl (adjusted odds ratio = 4.7, 95% confidence interval = 2.6-8.8, p < 0.01).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this study, we found that residence in a particular San Francisco zip code was associated with a substantially lower risk of PCP. This rather surprising result persisted even after controlling for other known risk factors for PCP such as PCP prophylaxis use, antiretroviral therapy use, CD4 cell count, and HIV RNA level.

Although this finding is unexpected, we feel that it is valid for several reasons. First, the protective effect of this zip code remained in all of the multivariate models analyzed. In addition, the odds ratio of 0.2 suggests that this result is unlikely to be secondary to chance. The PCP and control groups differed in use of PCP prophylaxis and antiretroviral therapy, with PCP subjects being less likely to use prophylaxis and antiretroviral therapy than control subjects. PCP subjects also had a lower mean CD4 cell count than control subjects. However, controlling for these variables actually strengthened rather than weakened the geographic finding. Because little is understood about the epidemiology and transmission of PCP, our data raise interesting questions about factors that lead to infection with Pneumocystis.

Few prior studies have examined P. carinii from a geographic viewpoint. Several studies have reported discovery of P. carinii DNA in water and air samples, suggesting that there may be environmental reservoirs (11). Recent data on molecular analysis of P. carinii specimens found differences in the prevalence of specific P. carinii strains in different cities and demonstrated that these differences were more closely associated with place of diagnosis rather than place of birth (23). However, no studies except the current one and a study by Dohn and colleagues have looked for geographic variations within cities (24). Fungi commonly inhabit the soil, and outbreaks of fungal infections such as histoplasmosis, coccidioidomycosis, and blastomycosis have been linked to places of soil disturbance (25). A recent study by Navin and colleagues found that patients with PCP were more likely to have a history of recent gardening or hiking and camping than matched controls (odds ratio 5.38, 95% confidence interval 1.39-20.8 for gardening; odds ratio 7.68, 95% confidence interval 1.34-44.1 for hiking/camping) (16). The protective zip code in our study, 94103, is an urban area of San Francisco. Many of the other zip codes associated with a higher risk of PCP are more residential and contain parks and small yards. Since the 94103 zip code represents one of the poorest areas of the city, it is possible that its residents have limited access to activities such as camping. Unfortunately, the retrospective nature of our study makes this information impossible to obtain. Lack of soil exposure in the 94103 zip code may be one explanation of the reduction in risk of P. carinii in this area.

The question of person-to-person transmission of PCP has been raised since the first reports of outbreaks in immunocompromised patients (6). Communicable diseases such as tuberculosis are more prevalent where the poverty rate is high and where group housing is common (17, 19). Finding a protective effect for PCP in one of these areas makes a case against person-to-person transmission of Pneumocystis. However, due to the retrospective nature of our study and the broad area covered by each zip code, we do not have enough information to reach a conclusion either way. Person-to-person transmission could be occurring if people living in other zip codes are more likely to have roommates than subjects in the 94103 area. Furthermore, although we know the addresses of the patients, this information does not necessarily reflect where the patients spent most of their time or with whom they had contact. More data on specific living situations, presence of roommates, and social contacts are needed before any theory of person-to-person transmission could be supported or disproved.

Past research has shown that being black is protective against PCP after adjustment for PCP prophylaxis use and CD4 cell count (29). These results were obtained from the Pulmonary Complications of HIV Infection Study and included a large number of patients from San Francisco. Our findings raise interesting questions about this earlier result. The zip code 94103 includes a significant proportion of the black population in San Francisco (12% of population, third highest rate in San Francisco). In our multivariate model adjusting for zip code, the race effect on probability of PCP was no longer present. It is possible that the racial findings of Stansell and colleagues may partially be due to a geographic effect or from other issues such as access to care or organism exposure as previously demonstrated by Hu and colleagues (30).

Injection drug use (IDU) is also common in the 94103 zip code. HIV-infected individuals who use injection drugs are at increased risk of bacterial pneumonia and tuberculosis (31, 32); therefore, the lower incidence of PCP in the 94103 zip code might be secondary to a higher incidence of bacterial pneumonia and TB. However, our analysis adjusted for IDU, and examination of other diagnoses established in the control patients did not show differences in the rates of bacterial pneumonia or TB among the zip codes.

Another intriguing aspect of these results is the similarity to a study by Dohn and colleagues (24). These authors report a geographic clustering of PCP cases within Cincinnati. Although they describe an increase in PCP in certain areas as opposed to the decrease seen here, the concept of varying risk for PCP according to geographic location is similar to the results reported in this study. Despite using different methods of analysis and studying different cities at different times in the AIDS epidemic, the conclusions reached are the same. Interestingly, Dohn found higher rates of PCP in relatively higher income areas in the city, an observation that is reproduced in our study. It is unexpected that lower rates of PCP would be found in poorer areas of the cities, suggesting that there may be some undiscovered factor that is associated with both poverty and the incidence of PCP.

The theories explaining our results are purely speculative. The presence of areas in San Francisco and Cincinnati associated with variable risk for PCP is an intriguing, but as yet unexplained, finding. Future studies performing more specific geographic analysis, examining soil samples and gathering data on soil exposure, or employing the techniques of molecular epidemiology may aid in elucidating the underlying cause of geographic variation in PCP risk. Information gained from such studies would provide valuable insights into the epidemiology and transmission of PCP and might lead to recommendations to avoid specific exposures or to place hospitalized patients with PCP in respiratory isolation and hopefully lead to a continued decline in the incidence of this significant disease.