Validating Use of Medical Record Data to Predict Risk Factors of Hospital-Acquired Pressure Ulcers

Purpose:
To derive and validate a model for predicting risk of hospital acquired pressure ulcers (HAPU) in adults, by day 5 of hospitalization.

Background/Significance:
HAPU are a serious patient safety issue. Current risk assessment scales for HAPU have limitations. It is unknown if factors found in the electronic medical record (EMR) are useful in predicting HAPU risk in a broad population of hospitalized adults.

Methods:
A retrospective and descriptive study conducted at a quaternary care medical center on EMR data abstracted on hospitalized med/surg adults during a 3-month period in 2009 (derivation model) and 2011 (validation model). Exclusions were age ≥90 years, BUN ≥ 60 mg/dL, creatinine ≥10 mg/dL, or hematocrit <10% or ≥50%. A full logistic regression model included all potentially important clinical factors. Data reduction was performed by choosing factors that best approximated the fit of the full model. Concordance index (CI) was used to discriminate HAPU development. Bootstrap resampling was performed to bias correct the estimate in the derivation and validation models.

Results:
The derivation and validation cohorts numbered 13,560 and 3,160 patients, respectively. By derivation, 12 factors were associated with HAPU; CI was 0.89. The validation cohort that used the same 12 factors had a CI of 0.78. Model factors performed better than chance at predicting HAPU; but calibration of low (<30%) HAPU risk was stronger than that of moderate or high HAPU risk. Models of Braden total/component scores performed worse than the 12-factor model in the derivation cohort (CI, 0.68), but similar to the 12-factor model in the validation cohort (CI, 0.81); however, calibration of predicted to actual HAPU was best when HAPU risk was below 20%.

Conclusions and Implications for Practice:
In hospitalized med/surg adults with multiple diagnoses, EMR factors were unreliable for predicting moderate-high risk of HAPU. New models need to be developed and tested that do not rely on EMR data; or that focus on specific populations or hospital environments.