Abstract
Objective:
To evaluate the impact of pre-intensive care unit admission (pre-ICU) statin use on all-cause in-hospital mortality and ICU length of stay (LOS).
Design:
Retrospective cohort study.
Setting:
Adult ICUs at tertiary hospitals.
Patients:
Adult critically ill patients diagnosed with sepsis admitted to the ICUs.
Intervention:
The exposure was pre-ICU statin prescription (statin users); unexposed represented absence of pre-ICU prescription (non-users).
Measurement and Main Results:
We used the 2001–2012 Medical Information Mart for Intensive Care-III (MIMIC-III) database to determine average treatment effect (ATE) of pre-ICU statin use on 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality using the Augmented Inverse Propensity Weighted technique (AIPW), after adjusting for confounding factors (age, race, health insurance, corticosteroids use, vital signs, laboratory tests, and Sequential Organ Failure Assessment score (SOFA). We measured 30-day ICU mortality as deaths within 30 days of admission to the ICU, and ICU LOS was measured in fractional days. 30-day in-hospital mortality was measured as death within 30 days of hospital admission. A total of 8,200 patients with sepsis were identified; 19.8% (1623) were statin users, and 80.2% (6577) were non-users. Most were Caucasian, aged 80 years and above, and male. After adjusting for confounding factors, pre-ICU statin use decreased 30-day ICU mortality (ATE, −0.026; 95% confidence interval [CI], −0.048 to −0.009); ICU LOS (ATE, −0.369; 95% Cl, −0.849 to −0.096); and 30-day in-hospital mortality (ATE, −0.039; 95% CI, −0.084 to −0.026) on average compared with non- statin use, respectively. In a stratified analysis, the result for ICU LOS (ATE, −0.526; 95% CI, −0.879 to −0.241) and 30-day in-hospital mortality (ATE, −0.023; 95% CI, −0.048 to −0.002) were consistent among patients admitted to the medical ICU.
Conclusions:
Among patients with sepsis admitted to the medical ICU, pre-ICU statin use is causally associated with a decrease in 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality compared to non-use. This study adds to the totality of evidence on the pleiotropic effect of statin use in patients with sepsis.
Keywords: Septic shock, Hydroxymethylglutaryl-CoA Reductase Inhibitors, Mortality, Length of Stay, Critical Care, Retrospective Studies, Organ Dysfunction Scores
Introduction
Sepsis is a major public health concern that affects about 1.7 million adults in the United States each year, contributing to more than 270 000 deaths1, 2 and accounting for more than $20 billion of total US hospital costs in 2013.3, 4 The incidence of sepsis and the number of sepsis-related deaths has been increasing each year.5 Severe sepsis and septic shock are major factors that contribute to mortality in critically ill hospitalized patients admitted to intensive care units (ICU).6, 7 According to the Centers for Disease Control and Prevention, sepsis is the leading cause of death in the ICU.8, 9 Sepsis is a highly complex inflammatory disease with different cell and humoral pathways, and there is a shortage of new therapies to enhance the survival of patients with sepsis.5
Most of the adjunctive therapies developed for sepsis over the past two decades were not successful in rigorous clinical trials.5 Therapies that attenuate inflammation may improve the outcomes in patients with sepsis.10 Initial randomized clinical trials (RCTs) of low-dose corticosteroids and recombinant human activated protein C showed promise for sepsis,6 yet subsequent pivotal RCTs and observational studies failed to confirm improvements in mortality and challenged their external validity.6,10 Several large real-world studies, a handful of small RCTs, and meta-analyses suggested that patients with sepsis or other life-threatening inflammatory conditions who received statins experienced improved clinical outcomes.11–13 In addition to their primary function as cholesterol-lowering medications, statins (including simvastatin, rosuvastatin, pravastatin, atorvastatin, fluvastatin, cerivastatin, pitavastatin, and lovastatin) have pleiotropic immunomodulatory and anti-inflammatory effects that are thought to be the mechanism that explains their effectiveness in patients with sepsis.9, 14
Despite the evidence of favorable outcomes of statins in patients with sepsis, other RCTs, observational studies, and meta-analyses suggest that statins do not improve health outcomes.15 Some researchers have argued that the protective associations reported in many of these studies could be explained by bias from “healthy user” effects in which statin users tend to have less severe comorbidities and better functional status than non-users and are more likely to practice other healthy behaviors.16–18 It is evident that the association of statins and the risk of mortality is inconclusive,15 and efforts to minimize all types of measurable confounding factors and other biases commonly encountered in observational studies using real-world data are needed. To our knowledge, no published observational study that examined the impact of pre-intensive care unit (pre-ICU) admission statin use on patients with sepsis used the relatively new method called the doubly robust Augmented Inverse Probability Weighting technique (AIPW) approach in accounting for confounding and minimizing biases.19 This approach was designed to attenuate selection biases arising from uncontrolled non-response and attrition, nonrandom treatment assignment in observational studies, and noncompliance in randomized experimental studies.19 Therefore, we aimed to evaluate the impact of pre-ICU statin use on all-cause in-hospital mortality and ICU length of stay (LOS) using the AIPW approach.
Material and Methods
Data Sources
This study is reported in line with the STrengthening the Reporting of OBservational Studies in Epidemiology (STROBE) statement.20 We conducted a retrospective cohort study of adult patients aged 18 years and above who were admitted to the ICUs (coronary care unit, cardiac surgery recovery unit, surgical intensive care unit, trauma/surgical intensive care unit, and medical intensive care unit) of Beth Israel Deaconess Medical Center in Boston, Massachusetts (BIDMC) between 2001 and 2012.21 We utilized the Medical Information Mart for Intensive Care-III (MIMIC-III) database, which is a large, de-identified, comprehensive critical care database of patients admitted to BIDMC ICUs.21 MIMIC-III is freely available data with patient’s bedside information such as vital signs, laboratory data, prescriptions and medical charts, and procedural and diagnostic codes (International Classification of Diseases, Ninth Revision (ICD-9-CM) codes).21 Data contain about 38,597 distinct adult patients with 49,785 admissions.21 The Laboratory for Computational Physiology at Massachusetts Institute of Technology utilized structured data cleansing and date shifting to de-identified data in accordance with the Health Insurance Portability and Accountability Act (HIPAA) standards.21
Identification of Sepsis Patients Cohort
Sepsis was defined in this study based on Angus abstraction criteria implementing the 2001 international consensus conference definition of severe sepsis using a set of ICD9-CM codes. The abstraction process followed two steps as described in Horng et al and Iwashyna et al.22, 23 First, we utilized a set of codes to identify infection and a set of acute organ dysfunction procedure codes to identify acute organ dysfunction. Based on these steps, we retrospectively identified patients with severe sepsis. We reviewed and updated the original Angus abstraction criteria and included additional new ICD-9-CM codes for sepsis, severe sepsis, septic shock, and codes indicative of viral meningitis, cholangitis, and orbital cellulitis.22, 23 We identified a total of 11,224 patients diagnosed with sepsis on admission into the ICU between 2001 and 2012. In cases where patients had multiple sepsis episodes, only the first sepsis episode was considered. Patients with zero (0) Sequential Organ Failure Assessment (SOFA) score and patients less than 18 years of age were excluded. Patients with missing data or without documented laboratory tests and vital signs taken within the first 24 hours of ICU admission were excluded. Average vital signs or laboratory tests were used in the case of multiple records in the first 24 hours. A total of 8,200 patients with sepsis admitted in the ICU met the inclusion criteria and were used as the analytical sample for this study.24
Medication Exposure
We used the MIMIC-III v1.4 prescription drug file to identify medication exposure based on generic and brand names. To evaluate the impact of statin use on 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality, we categorized exposure status into statin users and non-users. Statin users include patients with record of prescription for atorvastatin, fluvastatin, lovastatin, pitavastatin, pravastatin, rosuvastatin, or simvastatin before ICU admission. We defined non-users as patients without pre-ICU prescription records for the above statins.
Outcome Measures
The two primary outcomes of interest are ICU LOS and 30-day ICU mortality, and a secondary outcome was 30-day all-cause in-hospital mortality. Date of death (“DOS”), “INTIME” (ICU admission date and time) and “OUTTIME” (ICU discharge date and time) variables were used to determined mortality within 30-days of admission in the ICU. Only patients with records of DOS between the “INTIME” and “OUTIME” were included for 30-day ICU mortality analysis. ICU LOS measured in fractional days is included in the ‘ICUSTAYS’ file where it is defined as “the length of stay for the patient for the given ICU stay.” 30-day in-hospital mortality was determined using the DOS in the hospital “DEATHTIME”, date of admission “ADMITTIME” and date of discharge “DISCHTIME” from the MIMIC v1.4 admission file to determine mortality within 30-days of hospital admission.
Covariates
Covariates included for adjustment were demographic characteristics such as age, gender, race category, and health insurance. Clinical characteristics included were vital signs, laboratory tests, body mass index (BMI), Charlson Comorbidity Index (CCI), and SOFA score. Severe sepsis and septic shock diagnostic codes were excluded in the calculation of CCI. Patient’s height measured during the hospital admission and the average of weight measured during the first 24 hours of patient’s first ICU stay were used to calculate the BMI measured in Kg/m2. Using the data obtained during the first 24 hours of a patient’s first ICU stay, we calculated the SOFA score by adapting with MIMIC v1.4 data with the SQL script retrieved from MIMIC GitHub, Inc website (https://github.com/MIT-LCP/mimic-code/blob/master/concepts/severityscores/sofa.sql). The following laboratory tests were included: hemoglobin, albumin, white blood cell, potassium, sodium, bicarbonate, total cholesterol, high density lipoproteins, low density lipoproteins, C-reactive protein, and lactate. Laboratory test parameters included in the SOFA score calculation were not included as a covariate in the multivariable modelling. Vital signs included as covariates were heart rates, systolic blood pressure, diastolic blood pressure, respiratory rates, and body temperature. Additionally, we added potential immunomodulatory medications such as steroids, which affect the health outcome of sepsis, as covariates for adjustment.25 ICU types were categorized into medical ICU, cardio-coronary ICU(coronary and cardiac surgery ICU), and surgical ICU (trauma and surgical ICU).
Statistical analysis
We presented baseline characteristics as frequencies and percentages for categorical variables, mean with standard deviation and median with interquartile range for continuous variable. In the multivariate analysis, we used a doubly robust estimation using augmented inverse probability weighting (AIPW) to estimate the average treatment effect (ATE) of statin prescription. We performed separate AIPW modelling for each of the three outcomes: 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality. AIPW combines the propensity score model and the model for the outcome variables to generate potential outcome means (POM) and ATEs.26
The AIPW estimation method generates a doubly robust ATE estimate which is unbiased even if one of the models is mis-specified.26 AIPW includes propensity score modeling, which has a balancing score property thus, implying that weighting ensures that measured patient characteristics are balanced between statin users and non-statin users as in a RCT.26 Balanced characteristics across users and non-users is required to establish a causal interpretation of the effect of statin. Characteristics balance is assessed using standardized differences such that characteristics are considered balanced across treatment and control groups if the weighted standardized difference is closer to zero (0) compared to the unweighted standardized differences.26 Furthermore, using AIPW modelling approach, we performed stratification analysis examining ATE within various ICU sub-types (medical, surgical, and cardio-coronary units).
In this study, the ATE is an estimate of the average causal effect of statin use on 30-day ICU mortality, ICU LOS, or 30-day in-hospital mortality and represents a calculated difference between the POM of statin users and non-users. ATEs were considered significant at an alpha level < 0.05.
Sensitivity analysis
To examine the robustness of our findings, we performed sensitivity analysis using patient-specific propensity scores matching (PS-matching) approach. PS-matching were estimated by fitting a multivariate logistic regression model predicting the odds of pre-ICU statin use including various baseline covariates. PS-matched study samples were created by matching on the propensity scores based on a 1:1 greedy matching algorithm. Using 0.3 cut-off, units were matched only if the difference in the logits of the propensity scores for pairs of units from the two groups is less than or equal to 0.3 times the pooled estimate of the standard deviation. Residual unmatched patients were excluded. Using the matched population, we estimated the association between pre-ICU statin use and 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality to assess the sensitivity of the AIPW results to estimates from PS-matching, an integral component of the AIPW approach. Therefore, we examined sensitivity of AIPW estimate by comparing estimates from PS-matching approach with focus on the direction of the point estimates. All data analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).
Ethical Considerations
This project (Pro00100865) was approved and determined to be exempt from Human Research Subject Regulations by the institutional review board of the University of South Carolina on June 17, 2020.
Results
Figure 1 shows the flow chart of the cohort selection process. From the MIMIC-III v1.4 dataset, we identified a total of 46,520 patients with sepsis admitted to the ICU between 2001 and 2012 from which 11,224 patients with first ICU admission for sepsis were selected. After excluding 989 patients with post-admission statin use, a total of 10,234 patients with first ICU admissions for sepsis were selected. We also excluded 42 patients younger than 18 years old, 1,315 patients with no laboratory data, 340 patients with no vital signs data, and 337 patients with no lactate data during the first 24 hours of ICU admission to obtain a final analytical sample of 8,200 patients, of which 1,623 were pre-ICU statin users and 6,577 were non-users.
Figure 1:
Flow chart on the cohort selection process
ICU: Intensive Care Unit; SOFA: Sequential Organ Failure Assessment; MIMIC-III v1.4: Medical Information Mart for Intensive Care III version 1.4.
Table 1 summarizes the baseline characteristics of the study population including internally generated weighted and unweighted standardized difference based on the AIPW. The weighted standardized differences were smaller and less than 0.10 in all baseline characteristics compared to the unweighted standardized difference for all included covariates. This means that with weighting based on AIPW, included covariates were balanced across statin users and non-users.
Table 1.
Baseline characteristics of sepsis patients by statin use
| Statin Prescription | ||||||
|---|---|---|---|---|---|---|
| User (N=1,623) | Non-user (N=6,577) | Standardized Difference | ||||
| Characteristics | N | % | N | % | Unweighted | Weighted |
| Demographic Characteristics | ||||||
| Age Group, years | ||||||
| 18–50 (Ref) | 93 | 5.73 | 1213 | 18.45 | ||
| 51–79 | 620 | 38.20 | 2982 | 30.14 | −0.0220 | −0.0015 |
| 80+ | 910 | 56.07 | 3381 | 51.41 | 0.1238 | −0.0001 |
| Gender | ||||||
| Female (Ref) | 649 | 40.02 | 2795 | 42.51 | ||
| Male | 974 | 60.01 | 3782 | 57.49 | −0.0117 | 0.0004 |
| Race | ||||||
| White (Ref) | 1176 | 72.47 | 4564 | 69.40 | ||
| Black | 145 | 8.92 | 538 | 8.19 | −0.0264 | 0.0029 |
| Others | 302 | 18.61 | 1475 | 22.41 | −0.0024 | −0.0040 |
| Insurance | ||||||
| Public (Ref) | 1326 | 81.71 | 4613 | 70.15 | ||
| Private | 296 | 18.29 | 1963 | 29.85 | −0.0285 | −0.0010 |
| Clinical Characteristics [Median (IQR)] | ||||||
| Steroids Prescription | ||||||
| No (Ref) | 412 | 25.40 | 2171 | 33.01 | ||
| Yes | 1211 | 74.60 | 4406 | 66.99 | 0.1201 | −0.0005 |
| Intensive Care Unit Types | ||||||
| Surgical ICU (Ref) | 243 | 15.01 | 1446 | 22.00 | ||
| Cardio-coronary ICU | 592 | 36.52 | 1058 | 16.10 | −0.1412 | 0.0000 |
| Medical ICU | 788 | 48.47 | 4073 | 61.90 | 0.0521 | −0.0008 |
| Laboratory Tests (Mmol/L) | ||||||
| Albumin | 2.7 (2.2–3.2) | 2.6 (2.2–3.1) | 0.0310 | −0.0025 | ||
| White Blood Cell | 14.2 (9.1–40) | 13 (7.8–20.5) | 0.0021 | −0.0760 | ||
| Hemoglobin | 10.3 (9.2–13) | 10.5 (9.1–11.9) | 0.0901 | 0.0030 | ||
| Sodium | 139.1 (135–140.9) | 138 (134.7–142) | 0.0112 | 0.0006 | ||
| Potassium | 4.3 (3.7–4.7) | 4.4 (3.9–5.0) | 0.0041 | −0.0230 | ||
| Bicarbonate | 20.5 (19–25) | 17.7 (15.0–21) | 0.0120 | 0.0063 | ||
| Lactate | 2.1 (1.54 −2.90) | 5.10 (3.91–7.45) | 0.0001 | −0.0016 | ||
| Total cholesterol level (mg/dl) | 167 (160.1–198) | 184 (169–210) | 0.0471 | −0.0014 | ||
| LDL level (mg/dl) | 94.9 (87.5–135) | 100 (82.6–147) | 0.0029 | −0.0921 | ||
| HDL Level (mg/dl) | 50.1 (43.9–70) | 47.1 (40–68) | 0.0187 | 0.0001 | ||
| C-Reactive Protein (mg/l) | 89 (31.1–131) | 101 (49.1–147) | 0.0981 | 0.0064 | ||
| Vital Signs | ||||||
| Heart Rate (bpm) | 91 (78–107) | 100 (83–113) | 0.0090 | −0.0050 | ||
| Systolic Blood Pressure (mm Hg) | 107 (98–112) | 104 (96–112) | 0.2310 | 0.0025 | ||
| Diastolic Blood Pressure (mm Hg) | 54 (48–62) | 52 (48–61.5) | 0.0310 | 0.0003 | ||
| Respiratory Rates (bpm) | 21 (18.8–24.1) | 23 (20–26) | 0.0020 | −0.7250 | ||
| Body Temperature (Celsius) | 37 (36–37.5) | 37 (36–37.5) | 0.0040 | −0.0425 | ||
| Body Mass Index (Kg/m2) | ||||||
| 24.1 (16 – 37.6) | 25.8 (15.1–38.2) | 0.0240 | −0.0011 | |||
| Charlson Comorbidity Index | ||||||
| 2 (0–8) | 2(0–10) | 0.0810 | 0.0040 | |||
| SOFA Score | ||||||
| 9 (5–12) | 12 (7–14) | 0.0410 | −0.0023 | |||
ICU: Intensive Care Unit; SOFA: The Sequential Organ Failure Assessment; IQR: Interquartile Range, Kg/M2: Kilogram/ Meter square; mmol/L: Millimoles/Liter; REF: Reference; LDL: Low Density lipoprotein,; HDL: High Density Lipoprotein; BPM: breaths per minute; mmHg: millimeter of mercury; mg/l: milligram per liter; mg/dl: milligrams per deciliter.
Table 2 presents the comparison of all unadjusted outcomes (30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality) among statin users and non-users. A total of 1369 patients died in the ICU, of which 304 patients were treated with statin before ICU admission. Among statin users, unadjusted 30-day ICU mortality was higher compared to non-statin users (20.6% vs. 18.8%, p=0.04, respectively). A total of 2417 patients had record of in-hospital death within 30 days of admission, of which 450 were treated with statin before admission into the ICU. 30-day in-hospital mortality varied among statin users and non-statin users; p=0.02. ICU LOS was lower among statin users compared with non-users (5.01 vs. 5.61 days; p=0.02). Crude assessment shows that non-users of statin had higher mean ICU LOS than statin users by 0.60 days on average.
Table 2.
Unadjusted impact of pre-ICU statin use on outcomes: 30-day ICU mortality, ICU length of stay, and 30-day in-hospital mortality.
| Statin Prescription | |||||
|---|---|---|---|---|---|
| Users (N=1,623) | Non-users (N=6,577) | P-Value | |||
| N | % | N | % | ||
| Primary Outcomes | |||||
| 30-day ICU Mortality | |||||
| No | 1173 | 79.41 | 4610 | 81.23 | 0.038 |
| Yes | 304 | 20.58 | 1065 | 18.76 | |
| ICU Length of Stay (Fractional Days) | |||||
| N | 1623 | 6,577 | 0.0192 | ||
| Mean | 5.01 | 5.61 | |||
| Std. Dev | 6.85 | 7.78 | |||
| Secondary Outcome | |||||
| 30-day In-hospital Mortality | |||||
| No | 1173 | 72.3 | 4610 | 70.1 | 0.017 |
| Yes | 450 | 27.7 | 1967 | 29.9 | |
ICU: Intensive Care Unit, N: Number, Std. Dev: Standard Deviation.
Considering full sample ATE estimation in Table 3, the ATE of statin on 30-day ICU mortality was −0.026 (95% confidence interval [CI], −0.048 to −0.009; p=0.0371), which indicates that pre-ICU statin use reduces the average 30-day ICU mortality by 8% compared to non-statin use after controlling for confounders. In the ICU LOS estimation, the ATE of −0.369 days (95% Cl, −0.849 to −0.096; p=0.0259) indicates that pre-ICU statin use reduces the average ICU LOS of patients with sepsis by 6% compared to non-use after controlling for confounders. In the 30-day in-hospital mortality estimation, the ATE of −0.039 (95% CI, −0.084 to −0.026; p=0.0006) indicates that pre-ICU statin use reduces 30-day in-hospital mortality by 14% compared to non-statin use after controlling for confounders.
Table 3.
Analysis of causal effect of pre-ICU statin on 30-day ICU mortality, ICU length of stay, and 30-day in-hospital mortality
| Full Sample Analysis | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Primary Outcomes | Secondary Outcome | ||||||||||||
| 30-Day ICU Mortality | ICU Length of Stay | 30-Day In-hospital Mortality | |||||||||||
| Estimate | Wald 95% CI | P | Estimate | Wald 95% CI | P | Estimate | Wald 95% CI | P | |||||
| Parameter | |||||||||||||
| POM | Statin Users | 0.313 | 0.295 | 0.408 | <.0001 | 5.359 | 4.808 | 5.662 | <.0001 | 0.246 | 0.232 | 0.290 | <.0001 |
| POM | Non-users | 0.339 | 0.358 | 0.384 | <.0001 | 5.728 | 5.475 | 5.769 | <.0001 | 0.285 | 0.273 | 0.292 | <.0001 |
| ATE | −0.026 | −0.048 | −0.009 | 0.0371* | −0.369 | −0.849 | −0.096 | 0.0259* | −0.039 | −0.084 | −0.026 | 0.0006* | |
| Percentage reduction in ATE (%) | 8.39 | 6.44 | 13.68 | ||||||||||
POM: Potential Outcome Means; ATE: Average Treatment Effect; CI: Confidence Interval; P: P-value. Each model controlled for: age group, race category, insurance, SOFA score: The Sequential Organ Failure Assessment, steroid use, laboratory tests, and vital signs.
Statistically significant at alpha significant level ≤ 0.05
Percentage reduction in ATE was calculated as: ATE % = (POM non-users – POM statin users) *100/POM non-users
Table 4 presents the results of stratified analyses by ICU types. The impact of pre-ICU statin use on 30-day ICU mortality was not different for medical ICU, surgical ICU, and cardio-coronary ICU; (p=0.0739, p=0.8884, and p=0.1379, respectively). Among patients admitted to the medical ICU, the ATE of −0.526 (95% CI, −0.879 to −0.241; p=0.0133) days indicates that pre-ICU statin use reduces ICU LOS of patients with sepsis by 10% compared to non-use after controlling for confounders. The effect on ICU LOS was not different in surgical and cardio-coronary ICU patients; (p=0.4321 and p=0.5896, respectively). Among patients admitted to the medical ICU, the ATE of −0.023 (95% CI, −0.048 to −0.002; p=0.0140) shows that pre-ICU statin use decreases 30-day in-hospital mortality by 8% compared to non-statin use after controlling for confounders. No difference was observed with 30-day in-hospital mortality for patients admitted to surgical and cardio-coronary ICUs; (p=0.6649 and p=0.0794, respectively).
Table 4.
Analysis of causal effect of pre-ICU statin on 30-day ICU mortality, ICU length of stay, and 30-day in-hospital mortality stratified by ICU type
| Primary Outcomes | Secondary Outcome | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 30-Day ICU Mortality | ICU Length of Stay | 30-Day In-hospital Mortality | |||||||||||
| Medical ICU | |||||||||||||
| Estimate | Wald 95% CI | P | Estimate | Wald 95% CI | P | Estimate | Wald 95% CI | P | |||||
| Parameter | |||||||||||||
| POM | Statin Users | 0.341 | 0.289 | 0.401 | <.0001 | 4.593 | 4.246 | 5.015 | <.0001 | 0.277 | 0.226 | 0.290 | <.0001 |
| POM | Non-users | 0.368 | 0.360 | 0.390 | <.0001 | 5.119 | 4.882 | 5.273 | <.0001 | 0.300 | 0.279 | 0.314 | <.0001 |
| ATE | −0.027 | −0.047 | 0.063 | 0.0739 | −0.526 | −0.879 | −0.241 | 0.0133* | −0.023 | −0.048 | −0.002 | 0.0140* | |
| Percentage reduction in ATE (%) | 7.33 | 10.28 | 7.67 | ||||||||||
| Surgical ICU | |||||||||||||
| Parameter | |||||||||||||
| POM | Statin Users | 0.362 | 0.231 | 0.383 | <.0001 | 6.758 | 5.600 | 7.941 | <.0001 | 0.250 | 0.201 | 0.321 | <.0001 |
| POM | Non-users | 0.370 | 0.291 | 0.359 | <.0001 | 7.389 | 6.924 | 7.823 | <.0001 | 0.279 | 0.254 | 0.295 | <.0001 |
| ATE | −0.008 | −0.072 | 0.044 | 0.8884 | −0.631 | −1.724 | 0.269 | 0.4321 | −0.029 | −0.011 | 0.037 | 0.6649 | |
| Percentage reduction in ATE (%) | 2.16 | 8.54 | 10.39 | ||||||||||
| Cardio-coronary ICU | |||||||||||||
| Parameter | |||||||||||||
| POM | Statin Users | 0.352 | 0.283 | 0.371 | <.0001 | 5.491 | 4.999 | 6.233 | <.0001 | 0.202 | 0.190 | 0.279 | <.0001 |
| POM | Non-users | 0.394 | 0.334 | 0.401 | <.0001 | 5.784 | 5.479 | 6.357 | <.0001 | 0.233 | 0.219 | 0.290 | <.0001 |
| ATE | −0.042 | −0.096 | 0.014 | 0.1379 | −0.293 | −1.058 | 0.455 | 0.5896 | −0.031 | −0.042 | 0.006 | 0.0794 | |
| Percentage reduction in ATE (%) | 10.66 | 5.07 | 13.30 | ||||||||||
POM: Potential Outcome Means; ATE: Average Treatment Effect; CI: Confidence Interval; P: P-value. Each model controlled for: age group, race category, insurance, SOFA score: The Sequential Organ Failure Assessment, steroid use, laboratory tests and vital signs.
Statistically significant at alpha significant level ≤ 0.05.
Percentage reduction in ATE was calculated as: ATE % = (POM non-users – POM statin users) *100/POM non-users
Sensitivity analysis
The sensitivity analysis indicates that the result of the PS matching approach was consistent with that of AIPW. For the 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality, estimates of PS matching and multivariate regressions were similar to estimates of AIPW in terms of direction as shown in the supplementary table.
Discussion
Prior research including real-world observation studies, RCTs, and meta-analyses showed that statin use among patients with sepsis or other life-threatening inflammatory conditions is associated with improved clinical outcomes.11, 27–29 However, similar contradictory studies (mostly RCTs) reported that statin use does not improve health outcomes.15, 30 It is therefore noteworthy that the evidence of the association between statin use and the risk of mortality from sepsis and other clinical outcomes are inconclusive,15 with discordant evidence between RCTs and observational studies. Therefore, use of an advanced analytical approach to minimize the influence of measurable confounding factors and other biases commonly encountered in observational studies are strongly recommended.
Consistent with previous studies that have reported positive outcomes,11, 13, 27–29 we found that pre-ICU statin use confers a protective effect on patients with sepsis. Statin use before ICU admission is associated with decreased ICU LOS compared to those not treated with statins. This effect is consistent among patients admitted to medical ICUs. When we evaluated the association of pre-ICU statin use with mortality, we also found that pre-ICU statin use was associated with lower 30-day ICU mortality and 30-day in-hospital mortality. The impact on 30-day in-hospital mortality is consistent among patients admitted to the medical ICU. Our finding is in tandem with a recent study that reported a lower 90-day mortality among critically ill patients who were treated with a statin before admission.13 The pleiotropic effect of statin therapy, particularly its anti-inflammatory effects in the pathophysiologic pathway of sepsis, could explain this study’s findings. We did not observe a difference with in-hospital mortality between statin users and non-users in surgical and cardio-coronary ICU patients probably because of relatively fewer number of patients in both settings compared to the medical ICU patients.
To account for selection bias, which is characteristic of observational studies, we utilized the AIPW approach,26 which ensures that subjects are pseudo-randomized across treatment and control groups as obtained in a typical RCT. In using AIPW, this study generates an unbiased ATE of pre-ICU statin use in terms of LOS and in-hospital mortality among patients with sepsis who were admitted to the ICUs. AIPW enables simultaneous modeling of both the propensity to assignment into treatment groups (unbiased assignment into treatment and control groups using the inverse probability weighting approach) and the outcomes modeling using multivariate regression adjustment, thus generating doubly robust and unbiased estimates of the ATEs of statin use.26
After minimizing the biases inherent with observational studies using AIPW, this study shows evidence of the beneficial effect of statin therapy among patients with sepsis, which contradicts evidence reported by most RCTs. Although the methodological difference between observational studies and RCTs has been reported as the reason for the divergent results, our study followed a pseudo-randomized quasi-experimental approach that effectively accounted for selection biases and supports findings reported in most observational studies.11–13 Findings from this study refute the previous claim by Majumdar et al that a healthy user effect explains why most observational studies reported a positive effect of statins in patients with sepsis.31 The use of the AIPW approach eliminates the possibility of healthy user effect since assignment into treatment and control groups were determined by chance.30 There are plausible scenarios that could explain why most RCTs could not detect evidence of the beneficial effect of statins among patients with sepsis. A critical review of these RCTs showed that sepsis diagnosis was underreported, and most studies could not provide additional information after more data requests, thereby resulting in the possibility of including none representative sample of patients treated with statins and being assessed for sepsis outcomes.30 Secondly, given the underreporting of infection, most of the RCTs may have been at risk of type I error because of lack of sufficient power to detect existing association between statins use and sepsis.30. Furthermore, from a clinical and methodological perspective, it is likely that RCTs and observational studies are measuring different outcomes. It should be noted that the AIPW approach used in this study is not superior to large RCTs with complete reporting of data, but it attenuates biases associated with other observational studies and noncompliance in RCTs.
This study includes both strengths and limitations. The main strengths of our study include the use of novel AIPW analytical approach to generate doubly robust unbiased estimates, which are the ATEs of statins on patients diagnosed with sepsis. Thus, this study is among few studies that report evidence of the positive effect of statin therapy on sepsis among patients admitted to the ICU. Given that our study is a single-center retrospective cohort study, generalizability of our study may be limited to patients admitted to various BIDMC centers in Massachusetts. Variable inclusion in this study is limited to only relevant variables available in the MIMIC-III data v1.4, thus our study may have omitted potential residual confounding factors not captured in the MIMIC-III data v1.4. Also, MIMIC-III data v1.4 lacks year variables for each observation’s admission records and thus we were unable to adjust for years of ICU admission. Evidence shown in this study applies to only two statins prescribed for the selected cohort of patients with sepsis, most of which included atorvastatin and a few simvastatin prescriptions. Finally, at the time of this analysis, MIMIC-III data v1.4 available were for the years 2001 to 2012. The exposure was defined simply as any statin use or no statin use prior to ICU admission. Future studies therefore should compare outcomes using individual statins.
Conclusions
In this study, we demonstrated that patients with sepsis admitted to the ICU could benefit from pre-ICU statin use. We found that pre-ICU statin use is causally associated with a decrease in 30-day ICU mortality, ICU LOS, and 30-day in-hospital mortality compared to non-use. Although further studies using more diverse ICU populations are needed to delineate the external validity of our findings, this study adds to the totality of evidence on the pleiotropic effect of statin use in patients with sepsis.
Supplementary Material
Acknowledgement
We are appreciative to the MIT Laboratory for Computational Physiology and collaborating research groups who through the grant support from the National Institute of Biomedical Imaging and Bioengineering (NIBIB) of the National Institutes of Health (NIH) under award numbers R01-EB001659 (2003–2013) and R01-EB017205 (2014–2018) made available the MIMIC-III data.
Footnotes
Disclosure: The authors declare no conflicts of interest.
Supplementary materials
We include supplemental digital table of the sensitivity analysis results. These results include estimates of logistic regression, negative binomial regression generated using propensity score matched sample.
REFERENCES
- 1.Rhee C, Dantes R, Epstein L, Murphy DJ, Seymour CW, Iwashyna TJ, et al. Incidence and Trends of Sepsis in US Hospitals Using Clinical vs Claims Data, 2009–2014. Jama. 2017;318(13):1241–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rhee C, Jones TM, Hamad Y, Pande A, Varon J, O’Brien C, et al. Prevalence, Underlying Causes, and Preventability of Sepsis-Associated Mortality in US Acute Care Hospitals. JAMA network open. 2019;2(2):e187571–e. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Paoli CJ, Reynolds MA, Sinha M, Gitlin M, Crouser E. Epidemiology and Costs of Sepsis in the United States-An Analysis Based on Timing of Diagnosis and Severity Level. Crit Care Med. 2018;46(12):1889–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Databases H Healthcare cost and utilization project (HCUP). HCUP Nationwide Inpatient Sample (NIS) Healthcare cost and Utilization Project (HCUP) 1988–2009: MD, Rockville; 2013. [Google Scholar]
- 5.Kopterides P, Falagas ME. Statins for sepsis: a critical and updated review. Clinical Microbiology and Infection. 2009;15(4):325–34. [DOI] [PubMed] [Google Scholar]
- 6.Kaukonen KM, Bailey M, Suzuki S, Pilcher D, Bellomo R. Mortality related to severe sepsis and septic shock among critically ill patients in Australia and New Zealand, 2000–2012. Jama. 2014;311(13):1308–16. [DOI] [PubMed] [Google Scholar]
- 7.Vincent JL, Marshall JC, Namendys-Silva SA, François B, Martin-Loeches I, Lipman J, et al. Assessment of the worldwide burden of critical illness: the intensive care over nations (ICON) audit. The Lancet Respiratory medicine. 2014;2(5):380–6. [DOI] [PubMed] [Google Scholar]
- 8.Mayr FB, Yende S, Angus DC. Epidemiology of severe sepsis. Virulence. 2014;5(1):4–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gupta R, Plantinga LC, Fink NE, Melamed ML, Coresh J, Fox CS, et al. Statin Use and Hospitalization for Sepsis in Patients With Chronic Kidney Disease. JAMA. 2007;297(13):1455–64. [DOI] [PubMed] [Google Scholar]
- 10.Bernard GR, Vincent JL, Laterre PF, LaRosa SP, Dhainaut JF, Lopez-Rodriguez A, et al. Efficacy and safety of recombinant human activated protein C for severe sepsis. The New England journal of medicine. 2001;344(10):699–709. [DOI] [PubMed] [Google Scholar]
- 11.Truwit JD, Bernard GR, Steingrub J, Matthay MA, Liu KD, Albertson TE, et al. Rosuvastatin for sepsis-associated acute respiratory distress syndrome. The New England journal of medicine. 2014;370(23):2191–200. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Lee M, Lee C, Lai C, Hsu T, Porta L, Lee M, et al. Preadmission statin use improves the outcome of less severe sepsis patients-a population-based propensity score matched cohort study. BJA: British Journal of Anaesthesia. 2017;119(4):645–54. [DOI] [PubMed] [Google Scholar]
- 13.Oh TK, Song I-A, Lee JH, Lim C, Jeon Y-T, Bae H-J, et al. Preadmission Statin Use and 90-day Mortality in the Critically IllA Retrospective Association Study. Anesthesiology: The Journal of the American Society of Anesthesiologists. 2019;131(2):315–27. [DOI] [PubMed] [Google Scholar]
- 14.Parihar SP, Guler R, Brombacher F. Statins: a viable candidate for host-directed therapy against infectious diseases. Nature reviews Immunology. 2019;19(2):104–17. [DOI] [PubMed] [Google Scholar]
- 15.Wan Y-D, Sun T-W, Kan Q-C, Guan F-X, Zhang S-G. Effect of statin therapy on mortality from infection and sepsis: a meta-analysis of randomized and observational studies. Crit Care. 2014;18(2):R71–R. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.van den Hoek HL, Bos WJW, de Boer A, van de Garde EMW. Statins and prevention of infections: systematic review and meta-analysis of data from large randomised placebo controlled trials. BMJ. 2011;343:d7281. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Shrank WH, Patrick AR, Brookhart MA. Healthy user and related biases in observational studies of preventive interventions: a primer for physicians. J Gen Intern Med. 2011;26(5):546–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Patrick AR, Shrank WH, Glynn RJ, Solomon DH, Dormuth CR, Avorn J, et al. The Association between Statin use and Outcomes Potentially Attributable to an Unhealthy Lifestyle in Older Adults. Value in Health. 2011;14(4):513–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Kang JDY, Schafer JL. Demystifying Double Robustness: A Comparison of Alternative Strategies for Estimating a Population Mean from Incomplete Data. Statist Sci. 2007;22(4):523–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Lachat C, Hawwash D, Ocké MC, Berg C, Forsum E, Hörnell A, et al. Strengthening the Reporting of Observational Studies in Epidemiology–nutritional epidemiology (STROBE‐nut): An extension of the STROBE statement. Nutrition bulletin. 2016;41(3):240–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Johnson AE, Pollard TJ, Shen L, Li-wei HL, Feng M, Ghassemi M, et al. MIMIC-III, a freely accessible critical care database. Scientific data. 2016;3:160035. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Horng S, Nathanson LA, Sontag DA, Shapiro NIJb. Evaluation of the Angus ICD9-CM Sepsis Abstraction Criteria. 2017:124289. [Google Scholar]
- 23.Iwashyna TJ, Odden A, Rohde J, Bonham C, Kuhn L, Malani P, et al. Identifying patients with severe sepsis using administrative claims: patient-level validation of the angus implementation of the international consensus conference definition of severe sepsis. 2014;52(6):e39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Critical care medicine. 2001;29(7):1303–10. [DOI] [PubMed] [Google Scholar]
- 25.Bascones-Martinez A, Mattila R, Gomez-Font R, Meurman JH. Immunomodulatory drugs: oral and systemic adverse effects. Medicina oral, patologia oral y cirugia bucal. 2014;19(1):e24–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Lamm M, Yung Y-F, editors. Estimating causal effects from observational data with the CAUSALTRT procedure. Proceedings of the SAS Global Forum 2017 Conference Cary, NC: SAS Institute Inc; http://supportsascom/resources/papers/proceedings17/SAS0374-2017pdf; 2017. [Google Scholar]
- 27.Lee MG, Lee CC, Lai CC, Hsu TC, Porta L, Lee M, et al. Preadmission statin use improves the outcome of less severe sepsis patients - a population-based propensity score matched cohort study. British journal of anaesthesia. 2017;119(4):645–54. [DOI] [PubMed] [Google Scholar]
- 28.Falagas ME, Makris GC, Matthaiou DK, Rafailidis PI. Statins for infection and sepsis: a systematic review of the clinical evidence. J Antimicrob Chemother. 2008;61(4):774–85. [DOI] [PubMed] [Google Scholar]
- 29.Janda S, Young A, Fitzgerald JM, Etminan M, Swiston J. The effect of statins on mortality from severe infections and sepsis: a systematic review and meta-analysis. J Crit Care. 2010;25(4):656.e7–22. [DOI] [PubMed] [Google Scholar]
- 30.van den Hoek HL, Bos WJ, de Boer A, van de Garde EM. Statins and prevention of infections: systematic review and meta-analysis of data from large randomised placebo controlled trials. Bmj. 2011;343:d7281. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Majumdar SR, McAlister FA, Eurich DT, Padwal RS, Marrie TJ. Statins and outcomes in patients admitted to hospital with community acquired pneumonia: population based prospective cohort study. Bmj. 2006;333(7576):999. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.