Critically ill obstetric and gynaecology patients : the development and validation of an outcome prediction model.
Introduction: Outcome prediction tools have the potential to provide significant adjunctive information for intensivists. Critically ill obstetric and gynaecology patients constitute a unique subset of the general ICU (intensive care unit) population yet, there exists no outcome prediction model developed specifically for these patients. Objectives: To evaluate the APACHE II score, prospectively develop and validate an outcome prediction model, evaluate organ failure (Organ Failure score and SOFA score) and review the SIRS (Systemic Inflammatory Response Syndrome) response in a cohort of critically ill obstetric and gynaecology patients. Design: A prospective study conducted over a 2 year period in the Surgical ICU at King Edward VIII Hospital, Durban. Institutional ethics approval was obtained. Patients were allocated to one of the following categories: Obstetric hypertensive group (Group I), Obstetric non-hypertensive group (Group II) and Gynaecology group (Group III). Group III was further subdivided into a pregnant (Group IIIa) and a non-pregnant group (Group IIIb). Data captured included demographic details, clinical assessment, investigations, treatment, variables required for calculating the APACHE II score, organ failure (OF) assessment, SIRS criteria and patient outcome. The APACHE II system, organ failure assessment and SIRS was evaluated in the entire patient subset. For the purpose of the outcome prediction model, the subset was divided into 2 groups: a development group and a validation group. STATA 7 software was utilised for data analysis. Results: The dataset comprised 260 inpatients. Obstetrics and gynaecology cases represented 18.5 % of the total ICU population (n=1408). The majority of the patients were young (mean age 27 ± 10.5 years). The mean ICU stay was 5.5 ± 7.9 days. The observed mortality for Groups I, II, III, IIIa and IIIb was 23.4%, 43.2%, 42.9%, 33.3% and 55.5% respectively. The mean APACHE II score was significantly higher in nonsurvivors compared to survivors for all patient subgroups (p< 0.0001). However the APACHE II system performed variably in each of the 3 groups. The area under the curve for the ROC curves in each of the 3 main subgroups varied from 0.81 to 0.94 for APACHE II. Groups IIIa and IIIb were too small to permit ROC curve analysis. Age, mean arterial pressure, respiratory rate, temperature, the Glasgow Coma Scale score and pH were identified as significant outcome predictors. Using these parameters an obstetric and gynaecology outcome prediction (OGOP) model was developed for Groups I, II and III. The area under the curve for the ROC curves in each of the subgroups was >0.9 for the OGOP Model. A predictive equation could not be developed for Groups IIIa and IIIb (due to a small number of admissions in these two groups.) Duration and the number of organ failures, correlated with outcome. The duration and number of organ failures associated with mortality differed for each group. Three OF exceeding 72 hours, 3 OF exceeding 48 hours and 3 OF equal to 48 hours were invariably fatal in Groups I, II and III/IIIa/IIIb respectively. SOFA scores were significantly higher in nonsurvivors compared to survivors (p<0.0001). A day one SOFA score equal to 18 (Group I), 15 (Group ll) and 13 (Group III, IIIa, IIIb) was also invariably fatal. A SIRS response was noted in 94.2% of the patient cohort (245/260). The SIRS response varied in the subgroups. Sterile shock and septic shock were associated with a high mortality rate. Groups IIIa and IIIb differed with respect to the mean age, duration of hospital and ICU stay and mortality rate. Although these subsets were numerically restricted (24 and 18 admissions respectively), the results suggest that the two subsets are distinctly different in nature. Comment: The OGOP model is easier to calculate and it is superior to the APACHE II System. It needs to be validated in other local and international units. Organ failure assessment as well as the SIRS response provides useful supplementary outcome information. Although current outcome prediction tools are not designed for individual application, continued research and refinement of the available tools, as well as the exploration of novel methods, may one day result in "near-perfect" prediction estimates and further broaden the scope of their utility.