Statin Use and Hospitalization for Sepsis in Patients With Chronic Kidney Disease FREE

Sepsis is a major cause of morbidity and mortality in patients who have chronic kidney disease and are receiving dialysis. Patients receiving dialysis have rates of sepsis that are significantly higher than those of patients without kidney disease, and dialysis patients have higher rates of mortality from sepsis.¹ In addition, rates of sepsis have been increasing during the last few decades.² Although studies have attempted to identify risk factors associated with sepsis in the general population³ and in patients with kidney disease,^4- 5 no preventive treatment has been identified.

Statins are cholesterol-reducing medications that have become a cornerstone of primary and secondary prevention of cardiovascular disease. In recent years, pleiotropic effects of statins have been highlighted.^6- 8 Statins have been found to reduce cardiovascular events regardless of baseline low-density lipoprotein (LDL) cholesterol level,⁹ and they have been shown to reduce inflammatory factors such as C-reactive protein, which is also associated with a reduction in the rate of cardiovascular events.^10- 11

Experimental evidence in animals has indicated that statins may prevent sepsis and modulate the severity of sepsis.^12- 14 Two smaller observational studies in humans have shown lower rates of sepsis-related mortality among patients being treated with statins and a reduction in the incidence and severity of sepsis among patients treated with statins.^15- 16 A recent larger, population-based cohort study found a reduction in the incidence of sepsis among patients who were prescribed a statin after being hospitalized for a cardiovascular event.¹⁷

Therefore, our aim was to assess the effect of treatment with statin medications on the rates of sepsis in a prospective cohort study of patients who had chronic kidney disease and were receiving dialysis. We hypothesized that statins would reduce the incidence of sepsis among these high-risk patients.

We conducted a national prospective cohort study of incident dialysis patients by examining the association of statin use with the occurrence of sepsis. The CHOICE (Choices for Healthy Outcomes in Caring for ESRD) study was initiated in 1995 to investigate treatment choices and outcomes of dialysis care. Eligibility criteria for enrollment included initiation of long-term outpatient dialysis in the preceding 3 months, ability to provide informed consent, age older than 17 years, and ability to speak English or Spanish. The Johns Hopkins School of Medicine institutional review board and the review boards for the clinical centers approved the study protocol.

From October 1995 to June 1998, 1041 participants from 19 states were enrolled at 81 dialysis clinics associated with Dialysis Clinic Inc (Nashville, Tenn; n = 923), New Haven CAPD (New Haven, Conn; n = 86), or Saint Raphael's Hospital (New Haven, Conn; n = 32). A specimen bank was established to store blood samples from the Dialysis Clinic Inc enrollees, and specimens were obtained for 895 (97.0%) of the Dialysis Clinic Inc participants.

Statin use at baseline was determined by review of dialysis clinic notes, hospital discharge summaries, and computerized order entry records.¹⁸ Statin medications used by patients included lovastatin, fluvastatin, pravastatin, simvastatin, and atorvastatin.

Data about patient demographics, health behaviors, work history, medical history, preparation for dialysis, and history of sepsis were collected from a baseline self-report questionnaire. Race/ethnicity was initially assessed in CHOICE because dialysis patient outcomes (including sepsis) were known to differ by race.⁴ Race/ethnicity was self-reported by patients, with the following investigator-defined categories: white not Hispanic, Hispanic, African American, Asian or Pacific Islander, and other. For the purposes of analyses and because of small numbers in certain groups, we categorized race as white, black, or other. Height and weight were obtained from the Centers for Medicare & Medicaid Medical Evidence Report (form 2728). Dialysis modality at baseline was defined as the modality at 4 weeks after study enrollment. Late referral was defined as a first visit with a nephrologist less than 4 months before onset of dialysis.¹⁹ Vascular access information was obtained through review of discharge summaries, dialysis flow sheets, and dialysis clinic progress notes.²⁰ Rates of calcitriol treatment were obtained from Medicare billing data through the United States Renal Data System. Frequency of sit-down rounds at the dialysis unit, a process measure of the quality of care associated with outcomes, was based on clinic surveys and categorized as monthly, more often than monthly, or less than monthly.²¹

Comorbidity was assessed using the Index of Coexistent Disease, an instrument that has been validated as a predictor of death in dialysis populations. The Index of Coexistent Disease score ranges from 0 to 3, with 3 as the highest severity level, and is a composite measure of both the presence and severity of different comorbid conditions.^22- 23 Individual comorbidities, except history of sepsis, were abstracted from dialysis unit records, hospital discharge summaries, medication lists, consultation notes, diagnostic imaging, and cardiac imaging reports, which were collected at each dialysis unit, photocopied, and sent to New England Medical Center for abstraction and scoring. Two dialysis nurses, with previous training and experience in using the Index of Coexistent Disease, reviewed and scored all records. The reliability of data abstraction and severity scoring was assessed with a masked recoding of 45 medical records. Interrater reliability for Index of Coexistent Disease score was high (κ = 0.93).

Baseline nonfasting venous blood specimens were routinely collected at the Dialysis Clinic Inc facilities just before a dialysis session. Laboratory values for albumin, creatinine, and hematocrit were obtained from monthly laboratory tests at dialysis clinics. For patients in the specimen bank, specimens were also sent overnight to the central laboratory, where they were stored at −80°C. More than 95% of samples were frozen within 48 hours of venipuncture. Total and LDL cholesterol levels were obtained from aliquoted samples sent to Quest Diagnostics (Baltimore, Md). C-reactive protein and interleukin 6 values were obtained from analysis at the University of Vermont, Colchester.

The observation period for each patient began on the date of enrollment and continued until January 1, 2005. The primary outcome was hospitalization for sepsis. We used the United States Renal Data System hospitalization data to determine the primary cause for each hospitalization. We defined hospitalization for sepsis as any hospitalization in which the primary International Classification of Diseases, Ninth Revision (ICD-9) code was 038.0 to 038.9 (septicemia) or 790.7 (bacteremia).³International Classification of Diseases coding information for sepsis has been used widely in other studies of sepsis.^2,17 The sensitivity and specificity of ICD-9 codes for sepsis have been evaluated by comparing ICD-9 information with medical record review, using clinical consensus definitions of sepsis. The sensitivity of ICD-9 codes was found to be 75.4% to 87.7%, depending on the specific ICD-9 codes used.²⁴ The positive predictive value of 038.x codes for sepsis was found to be 88.9% to 97.7%, depending on the clinical definition of sepsis.² Only episodes in which the primary cause of hospitalization was sepsis were included in our analysis to avoid including cases in which infection was acquired during hospitalization.

To assess the presence, direction, strength, and statistical significance of an association between statin use and sepsis, we performed 2 analyses: a multivariate Poisson regression analysis on the entire CHOICE cohort, comparing occurrence of sepsis in statin users vs all other patients; and an analysis comparing occurrence of sepsis in statin users vs a control group identified through propensity-score matching. For both analyses, we compared patients initially receiving a statin vs control patients who were not receiving a statin. Baseline characteristics of the study groups were compared with the χ² test for categorical variables and t test for continuous variables. Crude incidence rates of sepsis were calculated by dividing the total number of hospitalizations for sepsis by the cumulative time at risk for each group. Time at risk for sepsis started with enrollment or the Medicare eligibility date, if the patient was not Medicare eligible at the start of the study. Time during hospitalization was excluded from time at risk. Patients were censored at death or kidney transplantation.

In the first analysis, multivariate Poisson regression models were used to obtain incidence rate ratios adjusted for potential cofounders. Potential confounders were identified by assessing the relation between each variable and use of statins and risk of hospitalization for sepsis (P≤.20 for both associations). Also, clinically relevant variables that did not meet these statistical criteria were still included to derive a full nonparsimonious model. By these criteria, the confounders included in the models were demographic information (age, sex, race, education level, and employment status), dialysis modality, comorbidities (history of myocardial infarction, cerebrovascular disease or transient ischemic attack, peripheral vascular disease, diabetes mellitus, sepsis, and Index of Coexistent Disease score), and laboratory measures (albumin, total cholesterol, hematocrit, C-reactive protein, and interleukin 6). No pairs of covariates with correlations greater than or equal to 0.7 were added to the models to avoid issues of collinearity. Subgroup analyses by age, sex, dialysis modality, diabetic status, malignancy history, sepsis history, C-reactive protein, LDL, and comorbidity were also performed.

In the second analysis, we also calculated incidence rates of sepsis and the incidence rate ratios in a propensity-matched subcohort. Propensity score matching is an established method to address a major limitation of observational studies, namely, confounding by indication.^25- 26 A propensity score is calculated as the estimated probability from logistic regression of a patient's being assigned to a given intervention, in this case, a statin. The following variables were used in the regression model to derive this propensity score for each patient: age, sex, race, history of myocardial infarction, cerebrovascular accident, peripheral vascular disease, diabetes mellitus, total cholesterol level, baseline dialysis modality, education level, and employment status. Variables were chosen because they were a clinical indication for statin use, clinically related to statin use, or statistically associated with statin use. Total cholesterol was treated as a categorical variable by dividing total cholesterol levels into 2 groups, less than 200 and greater than or equal to 200 mg/dL. Of the initial 143 statin users, 107 had complete information on all variables necessary to generate propensity scores. Using a greedy algorithm with a 5-digit match, each of these patients was matched to a control patient who had a similar probability of receiving a statin.²⁷ The 5-digit algorithm proceeds sequentially through 5-, 4-, 3-, 2-, and 1-digit matches on possible controls' propensity scores for each case, and once a control is selected, it is out of the pool, making the algorithm “greedy.”

We also performed several sensitivity analyses to test the robustness of our results. First, we used multiple methods to control for the possibility that certain patients may have had a disproportionate risk of sepsis. We analyzed patients who had more than 1 hospitalization for sepsis according to their first hospitalization only, ie, if a patient had 2 or more sepsis-related hospitalizations, we changed their total count to 1 but kept the same at-risk time used in the primary analyses. Additionally, Cox proportional hazards models, with stratification by clinic and censoring at patient death, transplant, or last date of follow-up (December 31, 2004), were also used to examine time to first sepsis-related hospitalization. We also excluded patients with a history of sepsis from multivariate and propensity-matched analyses to test whether this affected our results. Second, although drug and alcohol abuse are risk factors for sepsis, we had limited data on these variables. For this reason, we identified the subset of patients for whom these variables were known and then performed sensitivity analyses including all of the potential confounders cited above, with and without the addition of drug and alcohol abuse to assess the effect of these variables. Third, there is experimental evidence that activated 1,25-dihydroxycholecalciferol may modulate the response to infection in endothelial cells.²⁸ We assessed whether treatment for secondary hyperparathyroidism was significantly different between the statin group and the control group, and we added baseline calcitriol treatment information to a multivariate model to test whether it would change the relationship between statin use and sepsis. Fourth, statin use may be a marker of high-quality care. We added quality-of-care markers for predialysis care (late referral to a nephrologist) and for postdialysis care (frequency of dialysis unit sit-down rounds) to a multivariate model to determine whether quality-of-care variables would affect the relationship between statin use and sepsis. Fifth, we accounted for overdispersion in the distribution of our count data by examining the results with negative binomial modeling. Sixth, because we did not have cholesterol values before statin treatment, we excluded cholesterol from the list of variables used to derive the propensity score and then reconstructed a new propensity-matched subcohort comparing the results with the original propensity-matched subcohort. Finally, adjustment in the full cohort analysis for the propensity score was also performed.

The possibility of confounding by clinic was controlled with fixed-effects modeling, clustered on the clinic, which accounted for within-clinic correlation and between-clinic differences in baseline incidence. A 2-sided P<.05 was used as the level of statistical significance for all tests. Statistical analyses were performed using Stata software, version 8.2 (StataCorp, College Station, Tex) and SAS system for Windows, version 9.1 (SAS Institute Inc, Cary, NC).

Almost 14% of patients were receiving a statin at baseline and 86% were not (Table 1). Patients receiving a statin were more likely to be white; less likely to have ever used street drugs; more likely to receive peritoneal dialysis; more likely to have a higher hematocrit level; less likely to have a late referral to a nephrologist; more likely to have a higher total and LDL cholesterol level; more likely to have a history of diabetes mellitus (or diabetes as an assigned cause of end-stage renal disease), myocardial infarction, cerebrovascular accident, or peripheral vascular disease; and more likely to have a history of sepsis. Statin users tended to consume less alcohol per week and had higher education levels and employment rates, although these characteristics were not statistically significant.

In the propensity-matched subcohort, the only variable that was significantly different between statin users and control patients was history of sepsis (Table 2), with statin users more likely to have a history of sepsis.

Among all 1041 patients, there were a total of 303 hospitalizations for sepsis during follow-up (mean, 3.4 years). The crude incidence rate of sepsis among statin users was 41 per 1000 patient-years exposed compared with 110 per 1000 patient-years exposed in the control group. The crude incidence rate ratio of sepsis was 0.37 (95% confidence interval [CI], 0.22-0.61) (Table 3) or 63% lower in the statin group compared with the control group. After adjustment for demographic characteristics, dialysis modality, comorbidities, and laboratory values, the risk of hospitalization for sepsis remained statistically significantly lower by 62% among statin users (Table 3, Model 3).

The protective association of statin use on rates of sepsis was present in nearly all subgroups analyzed (Table 4).

Among the 214 patients who were matched according to their propensity to have been prescribed a statin, there were 54 hospitalizations during follow-up. The relative risk of sepsis per 1000 patient-years was 0.24 (95% CI, 0.11-0.49) for statin users compared with nonusers (Table 3), or 76% lower among statin users. There was evidence of possible interactions between subgroup characteristics and statin use in the propensity-matched subcohort by age (P = .06), dialysis modality (P = .08), and C-reactive protein (P = .05).

First, we used multiple methods to address the possibility that certain patients may have had a disproportionate risk of sepsis. In the overall cohort, we found that 110 patients had 1 episode of sepsis, 25 patients had 2 episodes, and 30 patients had 3 or more episodes. Limiting the outcomes to include only the first hospitalization, we found that the incidence rate ratio of sepsis was 0.50 (95% CI, 0.26-0.99). Next, Cox models examining time to first sepsis hospitalization gave an adjusted relative hazard of 0.50 (95% CI, 0.25-0.98). When patients with a history of sepsis were excluded from the full multivariate model and the propensity-matched subcohort, the incidence rate ratios were 0.40 (95% CI, 0.22-0.73) and 0.24 (95% CI, 0.11-0.51), respectively. Second, we controlled for baseline reported street drug use and heavy alcohol use by adding these 2 variables to the full multivariate model within the subset of patients for whom these variables were known; incidence rate ratio was 0.31 (95% CI, 0.14-0.69). It was 0.30 (95% CI, 0.13-0.67) within this subset of patients, without the addition of drug and alcohol use variables, showing that rates of sepsis were not materially affected by these variables. Third, when calcitriol use was added to the full multivariate model, the incidence rate ratio was 0.27 (95% CI, 0.11-0.71). Fourth, when we added quality-of-care markers, including late referral to a nephrologist and clinic frequency of sit-down rounds, to the multivariate model, the incidence rate ratio was 0.29 (95% CI, 0.14-0.60). Fifth, the use of a negative binomial model to account for possible overdispersion gave an incidence rate ratio of 0.60 (95% CI, 0.26-0.98). Sixth, when we calculated propensity score without using cholesterol as a predictor, the incidence rate ratio was 0.24 (95% CI, 0.11-0.49). Last, adjustment for propensity scores in the entire cohort also yielded similar results (Table 3, Model 4).

In this national prospective cohort study of patients who had chronic kidney disease and began receiving dialysis, statin use was associated with a large and statistically significant reduction in the incidence of sepsis. After controlling for potential confounding in a multivariate model, this association remained strong and statistically significant. Additionally, the protective association of statins on the incidence of sepsis was present and even stronger in a propensity-matched subcohort analysis.

Several mechanisms may explain the observed protective effect of statins on the occurrence of sepsis. Statins are known to have immunomodulatory properties.^29- 32 Statins may regulate the immune response to infection, thereby minimizing the risk of clinical sepsis in patients with infections. In animal models of sepsis, animals pretreated with statins before induction of sepsis had higher rates of survival.^12,14 It is possible that statins reduce the production of cytokines or mitigate the vasodilatory response to cytokines that results in septic shock.¹³ Statins may also have direct antimicrobial effects, as shown by studies of the effects of statin medications on human immunodeficiency virus, Salmonella, and yeast growth.^33- 36 Indeed, the first statin was originally identified from a fungus, Penicillium citrinum, and it has been postulated that this fungal organism secretes a statin as a selective advantage to prevent replication of competing microorganisms that require cholesterol for growth.^37- 38

Three clinical studies in nonrenal disease cohorts have compared the rates of sepsis in patients treated with statins to those of control (or untreated) patients. In a retrospective cohort study of 388 patients with bacteremia, Liappis et al¹⁵ found a significant reduction in mortality among patients who were taking a statin at their hospitalization. Almog et al¹⁶ conducted a prospective cohort study of 361 patients hospitalized with an acute bacterial infection and found a significantly lower rate of severe sepsis and intensive care unit admission among statin users. More recently, a larger study of patients hospitalized for cardiovascular events found that treatment with a statin medication was associated with a lower incidence of sepsis during a mean follow-up of 2.2 years.¹⁷ A meta-analysis of 90 056 trial participants³⁹ showed that death from nonvascular causes was reduced in patients who were treated with cholesterol-lowering medications, but infectious causes were not examined separately. Specific to the dialysis population, the Deutsche Diabetes Dialyse Studie trial⁴⁰ of atorvastatin use in diabetic hemodialysis patients showed that fatal infection was not different between the placebo and atorvastatin groups; however, the nonfatal infection rate and sepsis in particular were not examined. A study recently under way, the Study of Heart and Renal Protection, may have the opportunity to examine the effects of statin use on the incidence and severity of sepsis in patients with chronic kidney disease.⁴¹

Our study has certain limitations. This was a prospective, observational study. Observational studies are limited by confounding by indication for treatment because of lack of randomization. We attempted to control for this issue in multiple ways. We considered and adjusted for a large number of potential confounders. Additionally, we compared outcomes in a propensity-matched subcohort of patients who had similar likelihoods of being prescribed a statin. We performed multiple sensitivity analyses to further test our results. After controlling for confounding through these multiple methods, we continued to find a strong and statistically significant protective association between statin use and the incidence of sepsis. Nonetheless, the possibility that unmeasured bias affected our results cannot be entirely excluded.

In particular, the risk of confounding by indication must be addressed because the relationship between cholesterol and mortality in dialysis patients is complex. Multiple studies have shown an inverse relationship between cholesterol and mortality for patients receiving dialysis.^42- 44 Statin use could be a marker for higher cholesterol levels. Recently, it has been shown that the presence of inflammation or malnutrition can explain the inverse relationship between cholesterol and mortality.⁴⁵ In our study, the markers of inflammation and malnutrition, namely, albumin, C-reactive protein, and interleukin 6, were similar between statin users and controls. Because the paradoxic association between cholesterol and mortality among dialysis patients can be explained by the presence of inflammation or malnutrition and because the statin and control groups in our study had similar markers of inflammation and malnutrition, it is unlikely that the protective association found among statin use and incidence of sepsis can be ascribed to higher cholesterol levels in the statin group.

Another limitation of our study is that most patient and treatment factors were assessed at baseline. For factors that may change over time, such as vascular access, medication prescriptions, and dialysis modality, this assessment timing may lead to misclassification. However, statin misclassification would have most likely led to a dilution of any true protective effect if patients in the control group were in fact treated with statins or if patients in the statin group discontinued their medication. Given that misclassification and patient crossover between statin and control groups tend to favor the null hypothesis, our findings may represent a conservative estimate of the protective effects of statins on rates of sepsis. We also attempted to control for changes in certain variables by assessing their distribution at various points. For example, we assessed whether the differences in vascular access for hemodialysis patients were statistically different between statin users and control patients at enrollment and at 6 months. We did not find any substantial differences.

Last, we relied on United States Renal Data System and Medicare data to determine our outcome. Although this method has limitations, administrative information to determine cause of hospitalization has been used widely in other observational studies.^2,17ICD-9 codes have been found to have acceptable sensitivity and specificity when compared with medical record review using clinical consensus definitions of sepsis as the gold standard.^2,24 However, use of ICD-9 coding information to determine the primary outcome may have resulted in underdetection of sepsis. This limitation would result in loss of statistical power.

In summary, treatment with a statin was associated with a substantial reduction in the incidence of sepsis among patients who had chronic kidney disease and received dialysis. This protective association remained strong and statistically significant after adjustment for potential confounding. To our knowledge, this is the first study to show a strong and significant effect of a medication administered long term on lower rates of sepsis among patients with chronic kidney disease. In light of the high rates of sepsis and sepsis-related mortality in such patients, these findings are important and warrant the examination of the prevention of sepsis as a potentially important benefit in a randomized trial.