Population
Pre-CPOE and post-CPOE patients did not differ in their demographic characteristics (table 3) except for length of stay. Post-CPOE patients stayed significantly longer on the ward (Mean: 2.5 days) than in pre-CPOE (Mean: 2.1 days, p = 0.005, t-test).
Table 3
Demographic comparison before and after CPOE
|
Categories
|
pre-CPOE
N = 500
|
post-CPOE
N = 500
|
p Value
|
Sex
|
|
|
0.75
|
Female
|
219 (43.8%)
|
214 (42.8%)
|
|
Male
|
281 (56.2%)
|
286 (57.2%)
|
|
Age (in categories)
|
|
|
0.444
|
Preterm
|
3 (0.6%)
|
0 (0%)
|
|
Neonates
|
19 (3.8%)
|
11 (2.2%)
|
|
Infants
|
161 (32.2%)
|
167 (33.4%)
|
|
Children
|
225 (45.0%)
|
222 (44.4%)
|
|
Adolescents
|
92 (18.4%)
|
100 (20.0%)
|
|
Age (mean)
|
years
|
years
|
0.549
|
|
5.86 ± 5.39
|
6.06 ± 5.39
|
|
Weight
|
kg
|
kg
|
0.774
|
|
23.33 ± 19.50
|
23.68 ± 19.02
|
|
Height
|
cm
|
cm
|
0.879
|
|
112.05 ± 37.56
|
112.59 ± 38.33
|
|
Body Surface
|
[m2]
|
[m2]
|
0.835
|
|
0.861 ± 0.497
|
0.871 ± 0.480
|
|
Diagnosis (mean)
|
quantity
|
quantity
|
0.285
|
|
3.12 ± 2.89
|
3.33 ± 3.19
|
|
Length of stay
|
days
|
days
|
0.005*
|
mean
|
2.08 ± 2.23
|
2.52 ± 2.71
|
|
median
|
1
|
2
|
|
* Indicates significant value
Drug prescriptions
A total of 5022 drug prescriptions for 1000 patients were analyzed: 2299 drug prescriptions pre-CPOE and 2723 post-CPOE. Patients were prescribed more drugs post-CPOE than pre-CPOE (Mean = 5.5 drugs, vs Mean = 4.6 drugs, p < 0.001).
Handwritten prescriptions were significantly reduced by the CPOE (38.9% vs. 0.5%, p < 0.001).
Number of errors and exclusion of error 5.2, severity A-D
A total of 2485 errors for all prescriptions was found, of which 1802 errors occurred pre-CPOE and 683 errors post-CPOE. Consequently, 78.4 errors per 100 prescriptions (95% CI: 76.2 – 80.6) were found before the introduction of the CPOE versus 25.1 errors per 100 prescriptions (95% CI: 23.0 – 27.1) after introduction of the CPOE (p < 0.001, t-test). This means that 69.2% (95% CI: 67.4 – 71.0) of all prescriptions pre-CPOE contained at least one error versus only 22.8% (95% CI: 21.2 – 24.5) of all prescriptions post-CPOE, which implies a significant reduction of errors (p < 0.001). In 100 admissions, 360 errors could be found pre-CPOE, vs 137 in post-CPOE (p < 0.001).
Table 4
Types of error overall and severity of PCNE error 5.2
|
Categories
|
pre-CPOE
N = 1807
|
post-CPOE
N = 690
|
PCNE primary domains
|
|
|
1 drug selection
|
102 (5.7%)
|
109 (16.0%)
|
2 drug form
|
10 (0.6%)
|
17 (2.5%)
|
3 dose selection
|
203 (11.3%)
|
210 (29.7%)
|
4 treatment duration
|
8 (0.4%)
|
5 (0.7%)
|
5 dispensing
|
1452 (80.6%)
|
305 (44.7%)
|
8 patient transfer related
|
4 (0.2%)
|
17 (2.5%)
|
9 other
|
23 (1.3%)
|
27 (4.0%)
|
Error 5.2 (Necessary information not provided or incorrect advice provided)
|
1432 (79.5%)
|
301 (44.1%)
|
NCCMERP capacity to cause error (A)
|
89 (4.9%)
|
2 (0.3%)
|
NCCMERP does not reach patient (B)
|
264 (14.7%)
|
59 (8.6%)
|
NCCMERP no harm (C + D)
|
879 (48.8%)
|
188 (27.5%)
|
NCCMERP temporary harm (E + F + H)
|
200 (11.1%)
|
52 (7.6%)
|
NCCMERP permanent harm (G)
|
0
|
0
|
NCCMERP death (I)
|
0
|
0
|
However, a closer look at the types of errors (see table 4) showed that the most frequent type (80.6% of all errors pre-CPOE and 44.7% of all errors post-CPOE) was in PCNE primary domain 5 (dispensing). More precisely 79.5% of all errors pre-CPOE and 44.1% of all errors post-CPOE were type 5.2 errors: “Necessary information not provided or incorrect advice provided”. Examples for this type of error pre-CPOE are: “Missing or incorrect information about the ROA”, “active ingredient missing” (only product name prescribed), or “drug form missing”. Post-CPOE the most frequent error 5.2 was due to the additional selection of a mode of administration (MOA) like buccal or lingual, where it was not appropriate. Most of these errors 5.2 were of minor severity (NCC MERP severity grade A-D). 68.4% of all errors pre-CPOE and 36.4% of all errors post-CPOE were errors 5.2 with severity A-D. These minor formal errors due to missing information are unlikely to result in any harm for the patient (see Methods, Prescribing error [20]). Therefore, we decided to exclude errors 5.2 with severity A-D and to include only errors 5.2 with severity E-G into our analysis.
After exclusion of error 5.2 severity A-D, we counted a remaining total of 1004 errors, which were 570 errors pre-CPOE and 434 errors post-CPOE ( table 5). The overall error rate of 24.8 errors per 100 prescriptions (95% CI: 23.0 – 26.6) before CPOE was reduced to 15.9 errors per 100 prescriptions (95% CI: 14.3 – 17.6) after introduction of the CPOE, which was still a significant reduction of the error rate (p < 0.001). Therefore 22.5% of all prescriptions before CPOE contained at least one error (95% CI: 20.9 – 24.1), whereas the error rate after CPOE was only 14.7% (95% CI: 13.2 – 16.1) (p < 0.001). The rate of errors per 100 admissions was significantly reduced from 114 errors / 100 admissions pre-CPOE to 87 errors / 100 admissions post-CPOE (p = 0.004).
Table 5
Prescribing errors without PCNE error 5.2, severity A-D (NCC MERP)
|
Categories
|
pre-CPOE
prescriptions N = 2299
|
post-CPOE
prescriptions N = 2723
|
p Value
|
Total errors
|
570
|
434
|
|
Errors / 100 prescriptions
|
24.8
|
15.9
|
< 0.001*
|
Prescriptions with at least one error
|
517 (22.5%)
|
399 (14.7%)
|
< 0.001*
|
Errors / 100 admissions
|
114
|
86.8
|
0.004*
|
* Indicates significant value
Type of errors
As table 6 shows, the most frequent primary domain of errors pre-CPOE was domain 5 (dispensing), while post-CPOE the most frequent was domain 3 (dosing errors). Type 5 errors decreased significantly from 9.6 errors / 100 prescriptions to 2.1 errors / 100 prescriptions (p < 0.001), whereas dosing errors (type 3) showed no significant change (pre-CPOE: 8.8 errors / 100 prescriptions, post-CPOE: 7.5 errors / 100 prescriptions, p = 0.088).
Table 6
Types of prescribing errors according to PCNE classification
|
PCNE prim. domain
|
pre-CPOE
[errors / 100 prescriptions]
|
post-CPOE
[errors / 100 prescriptions]
|
p-Value
|
1 drug selection
|
4.44
|
4.00
|
0.450
|
2 drug form
|
0.43
|
0.62
|
0.361
|
3 dose selection
|
8.83
|
7.46
|
0.088
|
4 treatment duration
|
0.35
|
0.18
|
0.226
|
5 dispensing
|
9.57
|
2.06
|
<0.001*
|
8 patient transfer related
|
0.17
|
0.62
|
0.010*
|
9 other
|
1.00
|
0.99
|
0.975
|
Total
|
24.79
|
15.93
|
|
* Indicates significant value
The second domain which showed a significant change from pre- to post-CPOE was PCNE 8 “patient transfer related”. These errors increased significantly from 0.2 errors / 100 prescriptions to 0.6 errors / 100 prescriptions (p = 0.010). Type 8 errors were coded when two valid prescriptions for the same patient and time were found in different media, e.g. one CPOE prescription and one prescription on a paper chart. The doubled prescriptions contained in some cases the same information, but sometimes slightly different dosages or instructions.
The other primary domains did not differ statistically significantly pre- vs. post-CPOE.
The PCNE causes are displayed in figure 1. Details can be found in supplement 3. The following causes showed significant reduction in the error rate per 100 prescriptions:
- 1.4 “Inappropriate duplication of therapeutic group or active ingredient” (2.0 to 0.7 errors / 100 prescriptions, p < 0.001)
- 3.3 “Dosage regimen not frequent enough” (0.9 to 0.3 errors / 100 prescriptions, p = 0.010)
- 3.4 “Dosage regimen too frequent” (1.4 to 0.6 errors / 100 prescriptions, p = 0.003)
- 5.1 “Prescribed drug not available” (0.9 to 0.2 errors / 100 prescriptions, p < 0.001)
- 5.2 “Necessary information not provided or incorrect advice provided” (8.7 to 1.9 errors / 100 prescriptions, p < 0.001).
All other PCNE causes showed no difference pre- versus post-CPOE.
Severity of errors
The overall severity of error decreased significantly after the introduction of the CPOE. Mean rank pre-CPOE was 521 vs. 478 post-CPOE (p = 0.003, Mann-Whitney-test). See table 7.
Table 7
Severity of prescribing errors according to adapted NCC MERP index
|
|
Severity NCCMERP
|
pre-CPOE
N = 570
|
post-CPOE
N =434
|
p-Value
|
capacity to cause error (A)
|
2 (0.4%)
|
16 (3.7%)
|
0.002*
|
does not reach patient (B)
|
27 (4.7%)
|
22 (5.1%)
|
0.195
|
no harm (C + D)
|
117 (20.5%)
|
109 (25.1%)
|
0.082
|
temporary harm (E + F + H)
|
422 (74.0%)
|
284 (65.4%)
|
<0.001*
|
permanent harm (G)
|
2 (0.4%)
|
3 (0.7%)
|
0.826
|
death (I)
|
0
|
0
|
|
* indicates significant value
|
|
|
|
Errors that could potentially lead to patient harm were frequent in both periods. Pre-CPOE 74.0% of all errors were labelled as “temporary harm possible” (NCC MERP E, F, H) and 0.4% as “permanent harm possible” (NCC MERP G). Post-CPOE 65.4% were labelled with “temporary harm” and 0.7% with “permanent harm”. This means therefore that pre-CPOE in 100 prescriptions 18.5 errors occurred that could potentially lead to patient harm, whereas the rate of potentially harmful errors was reduced to 10.5 error / 100 prescriptions post-CPOE.
Handwritten vs. electronically written prescriptions
Pre-CPOE we compared the error rates in handwritten and electronically written prescriptions. Handwritten prescriptions showed a significantly higher rate of errors (29.3 errors / 100 prescriptions) than electronically written prescriptions (21.9 errors / 100 prescriptions) (p = 0.002). 26.6% of handwritten prescriptions contained at least one error, while only 19.9% of electronically written orders contained an error.
Interrater Reliability
Cohen’s Kappa for the agreement on whether or not an error occurred in a prescription was 0.476. This implies a moderate interrater agreement [26]. The agreement on primary domains and causes showed perfect agreement with k = 1.000 but for severity of error a kappa of only 0.158 (slight agreement) was calculated.