Background: Hospital discharge summaries communicate information necessary
for continuing patient care. They are most commonly generated by voice dict
ation and are often of poor quality. The objective of this study was to com
pare discharge summaries created by voice dictation with those generated fr
om a clinical database.
Methods: A randomized clinical trial was performed in which discharge summa
ries for patients discharged from a general Internal medicine service at a
tertiary care teaching hospital in Ottawa were created by voice dictation (
151 patients) or from a database (142 patients). Patients had been admitted
between September 1996 and June 1997. The trial was preceded by a baseline
. cohort study in which all summaries were created by dictation. For the da
tabase group, information on forms completed by housestaff was entered into
a database and collated into a discharge summary. For the dictation group,
housestaff dictated narrative letters. The proportion of patients for whom
a summary was generated within 4 weeks of discharge was recorded. Physicia
ns receiving the summary rated its quality, completeness, organization and
timeliness on a 100-mm visual analogue scale. Housestaff preference was als
o determined.
Results: Patients in the database group and the dictation group were simila
r. A summary was much more likely to be generated within 4 weeks of dischar
ge for patients in the database group than for those in the dictation group
(113 [79.6%] v. 86 [57.0%]; p < 0.001). Summary quality was similar (mean
rating 72.7 [standard deviation (SD) 19.3] v; 74.9 [SD 16.6]), as were asse
ssments of completeness (73.4 [SD 19.8] v. 78.2 [SD 14.9]), organization (7
7.4 [SD 16.3] v. 79.3 [SD 17.2]) and timeliness (70.3 [SD 21.9] v. 66.2 [SD
25.6]). Many information items of interest were more likely to be included
in the database-generated summaries. The database system created summaries
faster and was preferred by housestaff. Dictated summaries in the baseline
and randomized studies were similar, which indicated that the control grou
p was not substantially different from the baseline cohort.
Interpretation: The database system significantly increased the likelihood
that a discharge summary was created. Housestaff preferred the database sys
tem for summary generation. Physicians thought that the quality of summarie
s generated by the 2 methods was similar. The use of computer databases to
create hospital discharge summaries is promising and merits further study a
nd refinement.