The results of our review suggest a benefit and potential role for adjunct TPE in the treatment of sepsis with multiple organ failure. Prospective trials on this topic are lacking, and our results are among the largest in a body of evidence largely built on individual case reports and series. The 25% absolute reduction in mortality meets statistical significance and strongly suggests clinical benefit. The overall mortality is high in our study, but consistent with historical rates when adjusted for severity of illness (73% - 95.2% based on admission APACHE II and SOFA scores). Patients in both arms of our review had multiple comorbid conditions that increased mortality risk independent of sepsis, including hypotension requiring multiple vasopressors, acute renal failure, and moderate ARDS. The results, and the effect of TPE on outcomes are very similar to those seen in the prospective trial performed by Busund in a similar patient population (16).
In addition, patients receiving TPE in our trial had improved SOFA and cardiac SOFA scores at 48 hours, indicating improved organ function and hemodynamics. While the predicted mortality based on SOFA scores may be overstated, trends in SOFA scores serve as valuable predictors of outcomes (37, 38, 39). Fortenberry and colleagues reported improvement in organ dysfunction (as reflected by changes in PELOD scores from baseline) and 28-day mortality in septic pediatric patients meeting similar criteria who received TPE (37). The favorable fluid balance seen in patients receiving TPE was also noteworthy. This finding may be explained by endothelial stabilization, leading to improved hemodynamics and less need for volume resuscitation. In a retrospective study, where no research labs were collected, this cannot be proven, and future prospective studies should consider evaluation of endothelial markers. Nevertheless, the improved hemodynamic profile, organ function, and favorable fluid balance are all associated with improved outcomes and encourage further studies (40, 41, 42, 43).
The coagulopathy of sepsis is quite complex, composed of platelet dysfunction and abnormalities of the coagulation cascade. Thrombocytopenia, DIC, and decreased ADAMTS-13 activity have all been associated with poor outcomes in sepsis (31, 33, 34, 44, 45). We believe that coagulopathy is often present clinically, prior to laboratory derangements, so absolute values were not used to determine candidacy for treatment at our institution, or to monitor response to treatment. Retrospective analysis of platelet count did demonstrate more favorable outcomes in patients with higher platelet counts and resolution of thrombocytopenia, while lower platelet counts and failure of platelet recovery were associated more commonly with death (Table 6). We did not routinely measure markers of the coagulation cascade except in patients on anticoagulation, so we were unable to assess these values in our review. Interestingly, baseline ADAMTS-13 levels were similar in all patients in both arms, suggesting a possible association with severity of illness, as suggested in prior reports with sepsis (8, 45). We did not measure serial levels, so it is unclear whether TPE helped to restore activity, and if so, if this restoration was associated with improved outcomes. While none of these findings affected the clinical treatment of patients in this review, the data may serve useful in future prospective trials.
Our study has limitations beyond those common to small, retrospective, single center studies. First, the difference in time zero in the two arms potentially introduces bias. In a retrospective study, the intervention is easily defined, but since the control group did not receive treatment, we had to define an arbitrary time zero. Time zero for the intervention group was defined as the time of documentation of the initial TPE completion (regardless of ICU admission date and time). For control patients, time zero was defined as the time of the first recorded ICU vital signs. To limit bias, patients were propensity matched based on age, gender, number of vasopressors, lactate levels, chronic comorbidities, and APACHE II and SOFA scores on ICU admission. If anything, we believe this bias may favor the control group since patients who received TPE were potentially in the ICU longer before time zero as compared to the control group. Furthermore, SOFA scores calculated at time zero were higher in the intervention group compared to the control group, predicting a higher mortality in this group (Table 2), (p = 0.001).
While our institution does have a sepsis protocol, individual variation exists among providers. This variability is unlikely to influence outcomes, as multiple trials have demonstrated no difference in mortality using various resuscitation strategies (46, 47). In addition, since both the control group and intervention group were cared for by the same providers during the same time frame, variation between groups should be similar.
The decision to utilize TPE was provider dependent and involved an interdisciplinary approach between the attending intensivist and Nephrologist. General guidelines were developed (Table 1), but screening did not occur, and TPE was not considered unless the attending intensivist felt that it might be beneficial. Therefore, some patients that may have been candidates for TPE were likely not considered for treatment and likely fell into our control group. Furthermore, meeting the criteria did not guarantee that TPE would be provided. Ultimately, the decision was made by the consulting Nephrologist on a case to case basis. A large majority of TPE for sepsis was prescribed by a small number of providers within both groups. This bias cannot be eliminated from a retrospective trial where providing the intervention is not randomized, but using clearly defined, objective inclusion and exclusion requirements allows for matching and statistical comparison.
Another limitation of our trial was the lack of uniformity in duration of treatment in the intervention group. While most patients received between one and five treatments (92.5%), no objective guidelines were established at our facility to standardize the duration of TPE. Of the three patients receiving more than 5 treatments, two had prolonged admissions and received two separate courses of TPE, with different inciting infections. The third received treatment until normalization of platelets based on provider preference. Efficacy and duration were most often guided by hemodynamic response and lactate clearance. Many providers stopped TPE after vasopressor needs resolved, while others preferred a standing order for 3 or 5 treatments. Lactic acid levels declined more rapidly in patients receiving TPE, and levels were lower at 24 hours in survivors in both groups (Tables 5 and 6). Whether additional treatments would further enhance lactate clearance and improve mortality cannot be determined, but should be a priority in future, prospective trials.
ICU and hospital lengths of stay were longer in the intervention group but may not be reflective of true morbidity or cost as the standard care group had more early deaths. Additionally, more patients receiving TPE required new start renal replacement therapy, but the mortality associated with this treatment was clinically less in the TPE group (48% vs 78%, p = 0.06). There was no difference in new need for RRT at discharge in survivors in our trial. A larger sample size and longer follow-up interval are needed to assess the true impact on morbidity, resources, and long-term system costs.
The retrospective design of the trial was not optimal for detecting adverse events associated with TPE. All patients in our study were, by definition, hemodynamically unstable. It is impossible to attribute hemodynamic instability to TPE or to exclude TPE as a contributing factor based on our review of documentation. There were no recorded complications attributed to temporary dialysis catheter placement and no TPE treatments were aborted for clinical deterioration. However, we cannot exclude other potential adverse events that were unable to be tracked or identified. The potential adverse effects of TPE are well documented (48) and a recent pilot study demonstrated the feasibility and safety of early TPE in a similar patient population, reporting no adverse events (17). Nevertheless, a prospective, randomized trial should serve to identify potential adverse events associated with TPE specifically in the adult sepsis population.
The results of our trial are encouraging but limited by design, and the results cannot be used to change existing standards for the treatment of sepsis. The information gained from our experience offer valuable information and should be used to assist with design of a multicenter, randomized, controlled trial to better assess this potentially useful intervention.