Rahman SH, et al. - Researchers sought to externally validate a linear regression model developed by Swenson et al. in their hysterectomy population. In addition, they further validated the model in a cohort including robotic-assisted cases. Predictive variables included in this model were pre-operative hematocrit (HCT), patient weight, estimated blood loss, intra-operative crystalloid volume, pre-operative platelet count, and hysterectomy route which predicted postoperative day one (POD1) HCT ±5 for 100% of subjects using an internal validation set. The Swenson model was identified to be reliable for predicting POD1 HCT in their internal validation set, but it did not perform as well in the hysterectomy population of this study. The model seems to have utility as a screening tool if projected HCT≥35.

Methods

• Researchers performed a retrospective cohort study; data was collected from benign hysterectomies from April 2014 through May 2016.
• They calculated predicted POD1 HCT using the Swenson equation and compared it to measured HCT values.
• The results were compared to Swenson’s results using Chi-square or Fisher’s Exact.
• This analysis was then restricted to those with actual POD1 HCT≤30, to see if the model performed accurately in this sub-group which may need intervention.
• A receiver operating characteristic (ROC) curve with Louden Index was generated to determine the best cut-point from the Swenson HCT projections for predicting actual HCT≤30.
• Furthermore, the Swenson predicted HCT was stratified into 4 ranges: <32, 32-35, 35-38, and >38.
• This stratification allowed assessing differential accuracy of the Swenson model across HCT ranges.

Results

• This study included 602 hysterectomies; 478 subjects showed all variables needed for the Swenson model and POD1 HCT for comparison.
• In this data, the Swenson model was significantly less accurate compared to their validation set with fewer subjects whose predicted HCT was accurate at different thresholds from ±1% through ±5% of actual HCT (all p<0.001).
• They noted only 76.8% as accurate within ±5%.
• ANOVA showed similar accuracy among different surgical routes (p=0.193).
• When HCT=36.2, a quadratic best-fit curve suggested accuracy was maximized.
• Researchers observed projected HCT of ±2.5% of actual, but it was noted to be deteriorated at higher and lower HCT values.
• Accuracy was worse with only 55.3% of predicted HCT values within ±5% when the analysis was restricted to those with POD1 HCT≤30.
• The Swenson equation showed more tendency to overestimate HCT giving false reassurance in this subset.
• For predicting an actual HCT≤30, ROC Curve analysis showed that the best Swenson cut-point was 31.9 with sensitivity=75.5% and specificity=64.0%.
• Predicted HCT was finally divided into 4 groups: <32, 32-35, 35-38, and >38.
• When predicted HCT was <32 (n=164), the model more frequently under-predicted HCT and was least accurate in the subset most likely to need intervention.
• In this study, 17.2% (~1 in 6) showed actual POD1 HCT≤30 when predicted HCT was 32-35 (n=192).
• The percentage who had actual POD1 HCT≤30 dropped to 8.2% when predicted HCT≥35 was used as a cut-off.
• They identified no subjects with actual HCT<24, making this a reasonable choice for screening for anemia.