Data source and study design
We performed a retrospective cohort study using data from the MIMIC-IV (version 1.0), which included two in-hospital database systems—a custom hospital-wide electronic health record (EHR) and ICU-specific clinical information—that contain the deidentified, comprehensive clinical data of patients admitted to the ICUs of Beth Israel Deaconess Medical Center in Boston, Massachusetts, from 2008 to 2019. An individual who has finished the collaborative institutional training initiative examination (certification number 38995627 for author Huang) can access the database.
Participants
There were 382278 individuals and 51150 patients admitted to the ICUs during the study period. Patients were eligible if they (1) were ≥18 years old, (2) met the definition of Sepsis 3.0 criteria, which was defined as a suspected infection combined with an acute increase in the Sequential Organ Failure Assessment (SOFA) score ≥2 [18], and (3) had a SIC score ≥4 (Table S1) within the first 24 h (h) after ICU admission.
The exclusion criteria were (1) multiple ICU admissions; (2) age<18 years, ICU stay < 24 h; (3) usage of heparin for dialysis or treatment, rather than for prophylactic use, or LMWH administration or warfarin treatment during the ICU stay; (4) pregnancy; (5) a history of embolism and thrombosis; (6) a history of heparin-induced thrombocytopaenia; (7) hepatic failure; (8) chronic kidney disease stage 5; and (9) malignant cancer.
Research procedures and definitions
Data were extracted from the MIMIC-IV database through Structured Query Language [19]. We used the methods described in previous studies to search this database (sepsis) and analyse the extracted patient data [11, 20]. For patients with multiple hospitalizations, only the first hospitalization was included. The initial baseline characteristics and laboratory results for the first day of ICU admission were collected, including age at the time of hospital admission, sex, weight, laboratory results (white blood cell [WBC] count, platelet count, haemoglobin, international normalized ratio [INR], partial thromboplastin time [PTT], and prothrombin time [PT]), vital signs (mean arterial pressure [MAP], heart rate, temperature, respiratory rate, and partial pressure of oxygen [PO2]), comorbidities (hypertension, diabetes, chronic heart disease, and chronic pulmonary disease), urine output, use of vasopressors, mechanical ventilation, renal replacement therapy (RRT), acute kidney injury (AKI) stage, SIC score, length of hospital stay, and length of ICU stay. Clinical severity scales, including the SOFA score and Simplified Acute Physiology Score II (SAPS II), were also extracted. The SOFA score was calculated within the first 24 h after ICU admission. AKI was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria [21]. Both urine output and creatinine levels during the first 24 h after ICU entry were used to define AKI stages.
The laboratory variables platelet count and INR were measured throughout the entire ICU stay. The chart time of measurement and physiological values were extracted from the database. For patients with multiple measurements, the lowest daily platelet count value and highest daily INR value were included in the analysis. None of the screening variables had more than 10% of the values missing (Table S2). Single imputation was performed for variables with fewer than 10% of the values missing [22].
Exposure and outcomes
Participants were categorized into two groups: the heparin group, comprising patients who received heparin subcutaneously at preventive doses within 24 h after ICU entry, and the control group, comprising patients who received no heparin on the first day. The primary endpoint was ICU mortality, which was defined as patient survival at the time of ICU discharge. The secondary endpoints were 7-day mortality, 14-day mortality, 28-day mortality, and hospital mortality.
Statistical analysis
Categorical variables are expressed as numbers or percentages. They were compared between the heparin and nonheparin groups with the Chi-square test or Fisher’s exact test as appropriate. Continuous variables are expressed as the mean (standard deviation) or median (interquartile range [IQR]) as appropriate.
Propensity score matching (PSM) was used to account for the baseline differences in the probability of receiving heparin [23]. The PSM measures the probability of a patient being treated with heparin. In PSM analysis, the heparin group received prophylactic heparin within 24 h after ICU entry. Patients in the treatment group were matched to untreated patients by nearest neighbour matching. The standardized mean difference (SMD) was calculated before and after matching to examine whether PSM reduced the differences in pretreatment covariates between the treatment and control groups. Finally, a Cox proportional hazards model was used to adjust for residual imbalance by including parameters with P<0.05 and potential confounding judged by clinical expertise.
Heparin treatment during the ICU stay was considered a time-dependent variable in the marginal structural Cox model (MSCM). Potential baseline confounders, such as age, sex, use of mechanical ventilation, RRT, vasopressor, and the SOFA and SAPS II scores, were obtained on day 1 after ICU admission. Platelet count and INR throughout the entire ICU stay were included in the model as time-varying confounding factors. The parameters of the MSCM could be estimated using inverse probability weighting (IPW) to correct for both confounding and forms of selection bias such as informative censoring [24]. Weighting each patient by IPW allowed the creation of two pseudopopulations that were similar with regard to baseline and time-dependent confounding factors and different in heparin exposure. Details of IPW and the R code for performing the MSCM can be found in the electronic supplemental material (ESM) S1. The IPW package was used for estimating inverse probability weights [25].
Stratification analysis was conducted to explore whether heparin administration and ICU mortality differed across the subgroups classified by SIC, AKI, mechanical ventilation, and use of vasopressors. Subgroup analysis also used a Cox model adjusted for all variables in the patient baseline information. We explored the potential for unmeasured confounding between heparin and ICU mortality by calculating E-values [26]. The E-values quantify the required magnitude of an unmeasured confounder to negate the observed association between heparin and ICU mortality. Several prespecified subgroup analyses were performed with the MSCM. Two-tailed P values less than 0.05 were considered to indicate statistical significance. All statistical analyses were performed using the R package (version 4.1.1).