A systematic review of the clinical trials indexed in CENTRAL (PubMed), MEDLINE (OvidSP) and EMBASE (OvidSP) between 2001 and 2018 was performed. The search used the following keywords, which were expected to be present in the title and/or abstracts: "randomized studies" and "surgeries", and/or "perioperative", and/or "high-risk", and/or "complications", and/or "intraoperative" and/or "postoperative", and/or "cardiac output", and/or "cardiac index", and/or "hemodynamic monitoring", and/or "hemodynamic optimization therapy", and/or "hemodynamic intervention", and/or "cost-effectiveness", and/or "mortality". Methods identical to those recommended by the Cochrane systematic review of randomized controlled trials on increased blood flow to the organs, with explicitly defined goals and results after surgery, were used.
Only studies published in the English language were included. Two independent researchers identified the titles and abstracts of the potentially eligible studies. Disagreements between the investigators were resolved by consensus. Two other researchers extracted the following data from the full texts of potentially eligible studies: study design, patient population, interventions and outcomes. Similar to the approach used for the selection of texts, any disagreements between researchers regarding the data extraction were resolved by consensus. Each included study was assessed independently by the first and second reviewer for risk of bias in random sequencing generation, allocation concealment, blinding of participants and personnel, blinding of outcomes assessment, incomplete outcome data, selective reporting, and other sources of bias using the Cochrane Risk of Bias Tool.  In the absence of appropriate published data, at least one attempt was made to contact the authors of eligible studies to obtain necessary data. The analysis was performed with the best available information when there was no response.
The following information and outcomes were recorded: number of patients with respiratory complications (i.e., the need for respiratory support for more than 24 hours after surgery, hypoxemia, acute changes in lung mechanics), cardiovascular complications (need for hemodynamic support, such as the use of inotropes and vasopressors during the postoperative period), renal complications (oliguria, unexpected increase in creatinine and need for dialysis) and infectious complications (infections that occurred during the postoperative period) based on the records of each study. Mortality was assessed throughout the longer follow-up period (primary outcome) or was examined as in-hospital mortality. Many studies have reported complications as the number of complications rather than number of patients with complications, but we only reported the last unit in the analysis, which was the number of complications per patient.
To calculate the costs of these complications, only the length of ICU stay was considered.
The inclusion criteria of the selected studies were as follows:
1) studies in adult patients (18 years or older);
2) studies with patients undergoing hemodynamic optimization therapy with some type of cardiac output monitor;
3) studies that related the costs of hemodynamic optimization therapy and its outcomes, such as the reduction of mortality or morbidity rates or the reduction of length of hospital or ICU stay;
4) interventional studies comparing the use of invasive or minimally invasive monitoring with the standard strategy for hemodynamic optimization to alter clinical outcomes. The intervention should meet the following criteria:
Perioperative period: The administration of fluids with or without inotropes/vasoactive drugs to increase blood flow (standard therapy group) was compared with goals measured explicitly with invasive or minimally invasive hemodynamic monitoring (intervention group). The perioperative period started at the beginning of surgery and lasted up to 24 hours after surgery. Explicit goals were defined for the cardiac index, oxygen delivery (DO2), oxygen consumption, systolic volume, mixed or central venous oxygen saturation, oxygen extraction, or serum lactate.
The exclusion criteria were as follows:
1) animal studies;
2) studies published prior to 2001;
3) observational studies that did not use clinical intervention to change outcomes or case reports;
4) studies involving critically ill patients prior to intervention or with established sepsis who therefore with a higher probability of unfavorable outcomes and death regardless of intervention.
For the cost-effectiveness analysis, the costs were separated into 10 categories and two periods: the intraoperative period (monitoring and the costs of fluid infusion, inotropes or vasopressors and blood transfusions) and the postoperative period in the ICU, which was maintained at fixed daily rates regardless of disease, clinical exam and procedures depending on postoperative complications, laboratory diagnosis, use of antimicrobial and other agents (cardiac support, renal support, physical therapy and imaging). To avoid confounding factors, the costs of the surgical procedure (considering that the surgeries would have the same magnitude), postoperative analgesia and preoperative state were excluded from the final sum, as were costs related to the hospital infrastructure (electricity, safety system, etc.) and costs related to equipment depreciation, estate and management activities were not included. This approach makes different institutions comparable, not in terms of values but in terms of resource utilization. We believe that this simplified economic analysis can provide reliable and interchangeable data.
The overhead costs were estimated from a social perspective, i.e., regardless of who will bear the cost. However, the unit costs of health resources and services were obtained from the national databases of the Brazilian public health system, and therefore, the direct costs represent the costs borne by the payer [14-19].
The incremental cost effectiveness analysis was based on the difference in costs divided by the difference in survival days in each group.
The values of the results were described in two ways, in Brazilian currency (R$-BRL) and the equivalent in dollars ($-USD) for the present day.
The analyses were performed in Review Manager (RevMan 5.2.8) using fixed effects models with random effects models for comparison. In the fixed-effects regression methods the effect of interest is the same in all the studies and the differences observed between them are due only to sampling errors. While the random effects model assumes that the effect of interest is not the same in all studies, that is, the model incorporates a measure of the variability of effects between different studies. In this last model, the relation that the larger the sample size is, the greater the weight of the study in estimating the metanalytic measure.
We applied the intention-to-treat method for all analyses. Treatment effects were reported as relative risk (RR) and confidence intervals (95% CI) for clinical variables or as differences in weighted averages (SD) or medians for the length of ICU and hospital stays. Empty cells, the result of studies in which no event was observed in one or both arms, were corrected by adding a fixed value (0.5) to all cells with an initial value of zero. The chi-square test was used to assess whether the differences observed in the results were due to chance. A large chi-square (I2 statistic) provided evidence of heterogeneity of the intervention effects (indicating that the estimated effect was beyond chance).