An electronic documentation system increases diagnostic code capture for very low birth weight infants

Information describing medical interventions and patient outcome of very low birth weight (VLBW; BW <1500 grams) infants who receive neonatal intensive care is necessary to evaluate the quality of health care. We developed an electronic patient documentation system (EDS) called PCode to identify and select patient diagnostic (DX) codes and compared EDS to the previous paper-based documentation system (PDS). Our hypothesis was that EDS would capture more patient diagnostic codes than PDS.

PDS was originally developed as a two page 'code sheet' which listed all 275 neonatal diagnoses/management options and corresponding 3-4 character diagnostic (DX) codes, organized by organ system. PDS code sheets were manually completed by one of three trained 'coders' for every neonatal patient. EDS was developed as a Java application with the 'coder' selecting DX codes from drop lists also categorized by organ system. The study intervention was patient code selection using either the PDS or the EDS system. Both systems produced a paper DX code summary. PDS data capture occurred from 11/1997-3/1999 followed by EDS data capture until 8/2000. The primary outcome variable was DX codes captured per patient. Data were analyzed using independent t-test and linear regression.

The number of diagnoses increased with decreasing birth weight. After linear regression was applied to control for birth weight, the number of DX codes was 31% greater in the EDS group, 7.2 +/- 2.9 DX codes/patient, p=0.008. There was a trend towards fewer erroneous DX codes in the EDS group. The most common DX codes were hyaline membrane disease, sepsis evaluation and hyperbilirubinemia, as expected for this patient population.

EDS captured more VLBW infant DX codes compared to PDS. These results support the transition toward electronic data documentation for the neonatal clinical environment. Electronically linking PCode to the patient database has been initiated and should reduce manual coding errors.