Shock index thresholds to predict adverse outcomes in maternal hemorrhage and sepsis: A prospective cohort study

Introduction: Shock index (SI) is a predictor of hemodynamic compromise in obstetric patients. The SI threshold for action is not well understood. We aimed to evaluate SI thresholds as predictors of outcomes in obstetric patients. Material and methods: We undertook a prospective cohort study at three South African hospitals of women with postpartum hemorrhage (n = 283) or maternal sepsis (n = 126). The “first” and “worst” SI following diagnosis were recorded. SI was compared with conventional vital signs as predictors of outcomes. The performance of SI <.9, SI.9-1.69 and SI ≥1.7 to predict outcomes (maternal death; Critical Care Unit admission; major procedure; hysterectomy) and hemorrhage-specific outcomes (lowest hemoglobin <70 g/l; blood transfusion ≥4 IU) were evaluated. Results: “First” SI was one of two best performing vital signs for every outcome in postpartum hemorrhage and sepsis. In hemorrhage, risk of all outcomes increased with increasing “first” SI; for blood transfusion ≥4 IU odds ratio was 4.24 (95% confidence interval 1.25-14.36) for SI ≥1.7 vs SI.9-1.69. In sepsis, risk of all outcomes increased with increasing “worst” SI. Sensitivity, specificity, positive and negative predictive values of “first” SI <.9 vs SI ≥.9 for maternal death were 100.0%, 55.2%, 4.6% and 100.0%, respectively, in hemorrhage and 80.0%, 50.4%, 12.3% and 96.7%, respectively, in sepsis. Conclusions: The shock index was a consistent predictor of outcomes compared with conventional vital signs in postpartum hemorrhage and sepsis. SI <.9 performed well as a rule-out test and SI.9-1.69 and SI ≥1.7 indicated increased risk of all outcomes in both cohorts. These thresholds may alert to the need for urgent intervention and prevent maternal deaths.

This prospective cohort study was undertaken between January 2015 and May 2016 at three tertiary maternity units in South Africa (Groote Schuur, Tygerberg and Kimberley Hospitals). Women were eligible if they were diagnosed with either PPH or maternal sepsis (antepartum or postpartum) during their admission, up until discharge from hospital. PPH was defined as an estimated blood loss ≥500 mL for vaginal deliveries and ≥1000 mL for cesarean deliveries, as defined in studies from similar settings.15 Diagnosis of maternal sepsis was based on clinical features determined by the woman's healthcare provider and documented in the patient notes. Existing BP devices were replaced by the Microlife® CRADLE Vital Signs Alert (VSA), a vital signs device that measures BP and heart rate, is suitable for use in low‐resource settings16 and is validated as accurate for use in pregnancy, including preeclampsia and low BP in pregnancy.16, 17, 18 The device incorporates a traffic light early‐warning system triggering a green, yellow or red light according to categories of SI: SI <.9, SI .9‐1.69 and SI ≥1.7, respectively, to alert healthcare providers to abnormalities in vital signs. Anesthetic and recovery areas were allowed access to additional BP devices integrated into existing machines, if requested. Consequently, almost all women had access to the CRADLE VSA and traffic lights triggered by SI. Clinicians used the HR and BP measurements to determine clinical decisions and managed women according to local practice. They were not masked to the traffic light alerts but were not trained to escalate care according to them. The “first” and the “worst” sets of BP and HR taken following diagnosis of PPH or sepsis were recorded. The “first” set was defined as those documented immediately after diagnosis and the “worst” set as the set corresponding to the highest SI, documented at any time between diagnosis and discharge (or maternal death). The predefined outcomes included maternal death, maternal Critical Care Unit (CCU) admission, major surgical or invasive procedures and emergency hysterectomy. CCU was defined as a specified Critical Care area providing at least additional monitoring and interventions.19 Major surgical or invasive procedures were defined as uterine balloon tamponade, artery ligation/embolization/clamping, hemostatic brace suture, emergency laparotomy for hemorrhage, other procedures related to sepsis or emergency hysterectomy. For hemorrhage, additional outcomes included lowest hemoglobin following diagnosis <70 g/L and blood transfusion ≥4 IU. Perinatal complications were recorded but not treated as outcomes because, by definition, diagnosis of PPH and most sepsis diagnoses were made following delivery. We have previously demonstrated that SI is not influenced by regional anesthesia or mode of delivery, so this data was not collected.20 Data were extracted from patient notes reviewed by a local researcher. Data quality checks were undertaken on the database by an external researcher. Discrepancies were adjudicated by an obstetrician. Women with missing vital signs were excluded; women with missing outcomes were included but not analyzed for the outcome for which the data were missing. Predefined analysis aimed to determine whether SI was selected as a consistent predictor of adverse outcomes. This was calculated using area under the receiver operating characteristic curve (AUROC) values (95% confidence intervals [CI]) for “first” SI and conventional vital signs (HR, systolic BP, diastolic BP, mean arterial pressure, and pulse pressure) for predicting the predefined outcomes. Mean arterial pressure was defined as (2× diastolic BP + systolic BP)/3 and pulse pressure was defined as systolic BP − diastolic BP. Predictor equality of AUROCs across the outcomes was tested using unadjusted chi‐square analysis. The ability of the “first” and “worst” SI categories (SI <.9, SI .9‐1.69 and SI ≥1.7) to predict the risk of each outcome was evaluated using post‐test probabilities for each category, odds ratios of SI .9‐1.69 vs SI <.9 and SI ≥1.7 vs SI .9‐1.69, and non‐parametric trend testing of change in risk across the SI categories.21 Post‐test probability (with 95% confidence intervals) was defined as the proportion of women with vital signs falling within each category who have the outcome. The 95% CIs were included to allow for generalization from the sample to the population with similar characteristics. Post‐test probability was used to evaluate the performance of the three categories (SI <.9, SI .9‐1.69, and SI ≥1.7), rather than traditional predictive testing using sensitivity, specificity, positive predictive value and negative predictive testing, which is only appropriate when testing one threshold/two categories.22 Sensitivity, specificity, and positive predictive value and negative predictive testing were used to evaluate the test performance of the single threshold of SI <.9 as a rule‐out test at “first” vital signs measurement following diagnosis for the prediction of the two most severe adverse outcomes (emergency hysterectomy and maternal death). Separate analyses were performed for PPH and maternal sepsis. Women with both diagnoses were included in both groups. A post‐hoc power calculation was performed for two principal outcomes, CCU admission and emergency hysterectomy. The rate of CCU admission could be estimated to within 5.9% and 9.1% of the true value, for PPH and sepsis, respectively, with 95% confidence. The rate of emergency hysterectomy could be estimated to within 3.9% and 6.2% of the true value, for PPH and sepsis, respectively, with 95% confidence. Statistical analysis was performed in the statistical package STATA (version 11.2). The conventional significance level was set at P ≤ .05. The study was reported in accordance with STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines. The study was approved by the Stellenbosch University Ethics Committee (N14/06068, June 2014), University of Cape Town Ethics Committees (410/2014, July 2014) and the University of the Free State Ethics Committee (230408‐011, September 2014). Local ethics committees at two of the three sites (Tygerberg Hospital and Kimberley Hospital) required individual informed written consent to be obtained before the woman was enrolled in the study (or waiver of consent was granted if the woman was unconscious). Institutional‐level agreement for the study was given at the third site—Groote Schuur Hospital (ie, individual‐level consent was not required).

N/A

• nathan, hannah l.
• seed, paul t.
• hezelgrave, natasha l.
• de greeff, annemarie
• lawley, elodie
• anthony, john
• steyn, d. w.
• hall, david r.
• chappell, lucy c.
• shennan, andrew h.