Evaluation of a decision support system for initiation and
control of oral anticoagulation in a randomised trial
B Vadher, D L H Patterson, M
Leaning
Abstract
- Objectives: To determine whether a computerised
decision support system for initiation and control of oral
anticoagulant treatment improves quality of anticoagulant control
achieved by trainee doctors.
Design: Randomised controlled trial.
Setting: District general hospital in North London.
Subjects: 148 inpatients requiring start of warfarin
treatment.
Interventions: Management by trainee doctors (to
achieve therapeutic range of international normalised ratio of 2 to 3)
with indirect assistance from computerised decision support system
(intervention group) or without such assistance (control group).
Main outcome measures: Median time to therapeutic
range, stable dose, and first pseudoevent (excessive international
normalised ratio after therapeutic range has been reached) and person
time spent in the therapeutic range.
Results: 72 patients were randomised to the
intervention group and 76 to control group. Median time to reach
international normalised ratio of £2 was not significantly different
in the two groups (3 days). Median time to achieve a stable dose was
significantly lower in intervention group than in controls (7 days
v 9 days, P=0.01) without excessive overtreatment or
undertreatment with anticoagulant. Patients in intervention group spent
greater proportion of time in therapeutic range, both as inpatients
(59% v 52%) and outpatients (64% v
51%).
Conclusion: The computerised decision support system
was safe and effective and improved the quality of initiation and
control of warfarin treatment by trainee doctors.
- Introduction
- The quality of antocoagulant control during the initiation and
maintenance of warfarin treatment is generally poor.1-5
This may be due to the complex pharmacology of warfarin, failure to
follow guidelines, and the inexperience of trainee doctors, who are
responsible for managing most inpatient and some outpatient warfarin
treatment in Britain. Poor control of anticoagulant treatment may lead
to increased morbidity and mortality, longer hospital stays, and
therefore increased healthcare costs.
Several methods for predicting the initiation and maintenance doses of
warfarin have been developed.6 Regression methods are
based on the dose-response relation at a given time after loading doses
and do not require computerisation. A modified version of such a method
allows "flexible" loading doses on the second and third days after
starting treatment and then predicts the maintenance dose on day 4
based on the international normalised ratio.7-8 However,
the method is inflexible in that a loading dose of other than 10 mg on day 1 cannot be used, maintenance doses greater than 8 mg
cannot be determined, it aims for a wide target range of international
normalised ratio (2 to 4), and it gives no guidance on adjusting doses
after day 4. A randomised controlled trial found no significant
improvement in oral anticoagulation with this regression
method.9 Regression and mathematical methods have also
been developed and evaluated for control of oral anticoagulation in
outpatients.10
Computerised pharmacokinetic and pharmacodynamic models that
describe the time course of warfarin dose-response relations have also
been developed and evaluated.6 11-12 These allow more
flexible loading doses, with determination of a maintenance dose of any
size and for any therapeutic range. They also allow for adjusting doses
beyond day 4. However, these methods still require a doctor to
determine the final dose and the optimal monitoring interval. The
methods are complex, time consuming, and require expertise in
pharmacokinetics and statistics, and they may therefore not be
acceptable to all doctors. Furthermore, studies have not yet shown that
such methods have any significant advantage over regression methods.
We felt the need for a simple method that could provide guidance on
both initiation and maintenance of anticoagulation and which would be
acceptable to all doctors. We report the findings of our randomised
controlled trial of a simple dosing and monitoring method based on
simple proportional-derivative control in engineering
terminology.13
- Patients and methods
- Computerised decision support system
Using a computerised pharmacokinetic and pharmacodynamic model and
computer simulation, we developed an initiation regimen aiming for a
therapeutic range of international normalised ratio of 2 to
3.14 We developed a flexible regimen so that any loading
dose between 5 mg and 10 mg could be used on day 1. The
induction doses on the first three days were based on the daily
international normalised ratio and the previous dose. Adjustments of
the maintenance dose for outpatients were based on the error (target
international normalised ratio minus actual international normalised
ratio) and the previous dose (simple proportional controller), which is
a modified version of a maintenance controller.15
Proportional-derivative controllers (daily and weekly controllers) were
necessary for adjusting the maintenance dose in the phase between
induction and control of the maintenance dose for outpatients. The
doses in these controllers were based on the error, the rate of change
of error, and the previous dose.
For safety reasons, the decision support system would not recommend a
dose when there was no measurement of the international normalised
ratio (all controllers), when there was no previous ratio but a change
of dose by a doctor (daily and weekly controllers), or when the ratio
was very low (maintenance controller) or very high (weekly and
maintenance controllers). The ratio was monitored daily on the first
three days, and then monitoring intervals were based on the error and
previous interval.
We determined the error settings in the proportional-derivative
controllers by using a combination of the computerised pharmacokinetic
and pharmacodynamic model, 57 test cases, and the judgment of
experienced clinicians. We determined the error settings for the
outpatient controller empirically, mainly from the experience gained
from using a similar system at the Whittington hospital over the past
eight years.
Study design
We undertook this study at the Whittington Hospital, which has
acute medical and surgical units. The study was approved by the local
ethics committee, and we obtained verbal informed consent from the
patients who entered the study.
Inpatients who required initiation of warfarin treatment were eligible
for inclusion in the study. The exclusion criteria were patients who
had recently taken more than two doses of warfarin, patients whose
maintenance dose of warfarin was known, patients taking an oral
anticoagulant other than warfarin, a patient's or treating doctor's
refusal to participate in the trial, and therapeutic ranges of the
international normalised ratio outside 2-3. We did not exclude patients
with known hazards - such as other drugs and medical conditions - that
affected the dose-response relation of warfarin or those that might
predispose to haemorrhage.
Before any baseline hazards were recorded, we used simple randomisation
with a table of random numbers16 to assign the patients to
management by doctors aided by the decision support system
(intervention group) or to management by doctors alone (control group).
The doctors were mainly trainee doctors (n=42 for inpatients). For
outpatient control, the intervention group was managed by a nurse
practitioner and six trainee doctors aided by the decision support
system, while the control group was managed by the same six trainee
doctors without the assistance of the decision support system. The
patients were unaware of which study group they belonged to.
Patients were followed up until they reached one of the predetermined
study end points: warfarin stopped because duration of treatment was
completed or the diagnosis had been revised; patient followed up
elsewhere; any event such as death, major haemorrhage, or
thromboembolism; warfarin stopped for longer than one week for major
procedures; change of therapeutic range; or end of study period.
Study conduct
Patients' blood samples for measuring the international
normalised ratio were usually taken between 9 am and 11 am,
and warfarin doses were usually taken at about 5 pm. The
international normalised ratio was determined with a Sysmex CA1000
optical density coagulometer used with a low opacity Manchester
thromboplastin (international sensitivity index 1-1.2). The laboratory
participates in the British external quality control
scheme.17
We provided all the doctors with guidelines on anticoagulation from the
Drugs and Therapeutics Bulletin.8 For the
doctors treating patients in the intervention group, we also provided
the computerised decision support system's suggestion for the next
warfarin dose and interval to the next measurement of the international
normalised ratio. These doctors could reject the decision support
system's recommendations if they thought they were inappropriate. All
the doctors were free to seek advice from any expert.
For each patient, the daily warfarin dose, international normalised
ratio measurement, any identifiable changes in hazards, and any major
event such as death, major haemorrhage, or thromboembolism were
recorded. All the patients were given the Department of Health's
anticoagulant treatment booklet, which gives simple information on the
problems of warfarin treatment.18
Outcome criteria
The main outcome criteria for the initiation of warfarin treatment
were the time to reach an international normalised ratio of £2, the
time to reach a stable dose (defined as the first dose that maintains
the ratio between 2 and 3 for three consecutive days after starting
treatment), and the time to first pseudoevent (ratio Less/Not Equal 1.5 or £5)
after the therapeutic range is reached.
The main outcome criteria for controlling the maintenance dose was the
quality of anticoagulant control in inpatients and outpatients,
measured by the person time spent at a stable international normalised
ratio, assuming the ratio changed linearly between two
measurements,19 and the frequency of measurements of the
ratio. We did not include major haemorrhagic or thromboembolic events
requiring hospital admission as major outcome measures because these
tend to occur infrequently and would have required a much larger
trial.
Statistical analysis
We analysed data from all the patients up to their study end point
and analysed data from the patients in the intervention group
regardless of whether the computerised decision support system's
advice was accepted, rejected, or not available (that is, on an
"intention to advise" basis). All "time to event" data were
analysed by Kaplan-Meier curves and the log rank test. We analysed the
quality of anticoagulant control by methods suggested for measuring
rates of recurrent events because individual patients would have had
unequal follow up and events might have recurred in an
individual.20-21 An event was taken as a day spent at a
particular international normalised ratio. The interval between
measurements of the international normalised ratio was analysed by the
Mann-Whitney test. The tests were two sided, and a P value of 0.05
was used as the level of significance.
-
Results
- Table 1 shows the patients' demographic and clinical
characteristics. Of 170 consecutive inpatients who were considered, 148
were randomised to treatment while the other 22 were excluded because
of failure to recruit them or because they satisfied one or more of the
exclusion criteria. Of the 76 patients in the control group, 64 were
followed up as outpatients, as were 53 of the 72 patients in the
intervention group (P=0.11). Duration of follow up of patients in the
two groups was not significantly different (P=0.69). Four patients died
during the study - three from metastatic carcinoma and one from bowel
obstruction. None was due to poor anticoagulant control. One
thromboembolic event in each study group was due to undertreatment with
anticoagulant (international normalised ratio 1.5 in control group and
1.9 in intervention group), while four haemorrhagic events were due to
overtreatment with anticoagulant - three in the control group (ratios
3.5, 4, 10) and one in the intervention group (ratio
10).
| Table 1 - Demographic and clinical characteristics of
148 hospital inpatients requiring oral anticoagulation by treatment
group* (values are numbers of patients unless stated otherwise) |
| Control group (n=76) | Intervention
group (n=72)
|
| Median (range) age
(years) | 65 (18-92) | 67
(15-93)
| Men:women | 40:36 | 31:41
| Diagnosis: |
| Deep vein
thrombosis | 42 | 29
| Pulmonary
embolus | 14 | 26
| Atrial
fibrillation | 15 | 12
| Valve
disease | 2 | 2
| Mural
thrombus | 0 | 1
| Systemic
embolus | 2 | 0
| Prophylaxis | 1 | 2
| Baseline hazards
present+ | 65 | 61
| Heparin given during
hospital stay: | 67 | 66
| Median (range)
maintenance dose (mg) | 5
(1-18) | 5 (1-13)
| Study end
points: |
| Stopped: |
| Duration
of treatment complete | 25 | 22
| Diagnosis reviewed | 2 | 3
| Operation | 2 | 4
| Event: |
| Death | 2 | 2
| Haemorrhage | 4 | 2
| Thromboembolism | 1 | 4
| Patient
moved away | 11 | 8
| End of
study | 27 | 26
| Change of therapeutic
range | 2 | 0
| Violation of
protocol | 0 | 1
| Median (range) length of
follow up (days) | 88
(5-389) | 93 (3-392) |
| *Intervention group was treated by doctors advised by
computerised decision support system, while control group was treated
by doctors alone.
+Anaemia (haemoglobin concentration >115 g/l); renal impairment
(serum urea >10 mmol/l and creatinine >130 µmol/l); liver
impairment (serum bilirubin >17 µmol/l and serum aspartate
transaminase >40 IU/l, serum gamma-glutamyl transferase >40 IU/l, or
serum albumin >30 g/l); abnormal baseline clotting tests; excess
alcohol intake (>42 units/week in men, >28 units/week in women). Some
patients had more than one baseline hazard. | | | | | | | | | | | | | | | | | | | | | | |
-
Initiation of warfarin treatment
Fig 1 - Kaplan-Meier curve for time for patients
requiring anticoagulation to reach therapeutic range (international
normalised ratio £2) after start of warfarin treatment (intervention
group was treated by doctors advised by decision support system, while
control group was treated by doctors alone)
Figure 1 shows the Kaplan-Meier curve for the time to achieve an
international normalised ratio of £2. One patient in the control
group and three patients in the intervention group were censored. Eight
patients in the control group and four patients in the intervention
group were below the therapeutic range on discharge from
hospital.
Fig 2 - Kaplan-Meier curve for time for patients
requiring anticoagulation to achieve a stable dose after start of
warfarin treatment (intervention group was treated by doctors advised
by decision support system, while control group was treated by doctors
alone)
Figure 2 shows the Kaplan-Meier curves for the time to achieve a stable
dose. Fourteen patients in the control group and 11 patients in the
intervention group did not achieve dose stability before reaching a
study end point.
Fig 3 - Kaplan-Meier curve for time to first
pseudoevent (international normalised ratio Less/Not Equal 1.5 or £5) after
therapeutic range is reached among patients requiring anticoagulation
(intervention group was treated by doctors advised by decision support
system, while control group was treated by doctors alone)
Figure 3 shows the Kaplan-Meier curves for the time to the first
pseudoevent. More pseudoevents occurred in the control group than the
intervention group (41 v 25). Eighteen of the
pseudoevents in the control group were due to overtreatment with
anticoagulant compared with 12 in the intervention group. The median
time to the first pseudoevent among inpatients (whose compliance and
drug treatment were known) was 8.7 (SE 2.32) days in the intervention
group and 7 (2.64) days in the control group
(P=0.03).
Control of maintenance dose of warfarin
To compare the quality of oral anticoagulant control, we examined
the international normalised ratios after the therapeutic range had
been attained. Table 2 shows the time spent at various ratios by the
two patient groups as inpatients and outpatients. The intervention
group spent more time within the therapeutic range than the control
group, both as inpatients (59 v 52 days) and outpatients
(64 v 51 days).
-
| Table 2 - Quality of anticoagulant control among
patients given warfarin as inpatients and outpatients by treatment
group* (values are days per 100 patient days of treatment) |
| Inpatients | Outpatients
| INR+ | Control group (n=62) | Intervention group
(n=60) | Relative rate
(95%
CI)‡ | Excess days (95% CI)§ | Control
group (n=64) | Intervention group
(n=53) | Relative rate (95%
CI)‡ | Excess days (95% CI)§
| >1.5 | 5.6 | 1.3 | 4.2
(2.2 to 7.9) | 4.3 (0 to
1.2) | 4.2 | 1.3 | 3.3
(1.3 to 8.7) | 2.9 (0.3 to 5.5)
| >2.0 | 21.4 | 18.3 | 1.2 (0.8
to 1.7) | 3.1 (-4.2 to
10.4) | 31.8 | 21.1 | 1.5
(1.1 to 2.1) | 10.7 (2.1 to 19.2)
| 2-3 | 52.2 | 59.4 | 0.9 (0.7 to
1.0) | -7.2 (-16.3 to
1.9) | 51.0 | 63.7 | 0.8
(0.7 to 0.9) | -12.7 (-21.6 to -3.8)
| >3.0 | 26.4 | 22.3 | 1.2 (0.8
to 1.7) | 4.1 (-4.3 to
12.6) | 17.2 | 15.1 | 1.1
(0.7 to 1.8) | 2.1 (-5.6 to 9.7)
| >5.0 | 2.8 | 1.2 | 2.4 (0.6 to
9.8) | 1.6 (-0.8 to
4.1) | 1.1 | 0.8 | 1.5
(0.2 to 14.3) | 0.3 (-1.5 to 2.2) |
| *Intervention group was treated by doctors advised by
computerised decision support system, while control group was treated
by doctors alone.
+International normalised ratio.
‡Comparing rate in control group with rate in intervention
group.
§No of days per 100 patient days of treatment spent at given
range of international normalised ratios in the control group in excess
of that in the intervention group.
Total follow up times for inpatients of control and intervention
groups were 571 and 596 days respectively, and for outpatients of
control and intervention groups were 6331 and 6032 days respectively. | | | | | | |
For inpatients, the median interval between tests was 2 (range
1-22) days in the intervention group and 2 (1-30) days in the
control group (P=0.07). For outpatients, the median interval between
tests was 14 (2-63) days in the intervention group and 14
(1-91) days in the control group (P=0.2). There was no difference
between the study groups in the attendance rate at outpatient
clinics.
- Discussion
- Initiating warfarin treatment
The therapeutic range was attained more quickly in the patients in
the intervention group. A higher proportion of patients in the control
group were not in the therapeutic range as outpatients, when they would
not have been receiving heparin and would therefore have been at high
risk of recurrent thromboembolism. A stable dose was determined more
quickly in the intervention group without excessive undertreatment or
overtreatment with anticoagulant, as shown by the time to the first
pseudoevent. Theoretically, this should allow earlier discharge from
hospital with less frequent monitoring.
Our results of the initiation of anticoagulation are similar to those
in other randomised controlled trials comparing computer assisted
control with control by doctors only.11 22-23 White
et al found that the mean time to attaining the
therapeutic range and a stable dose was 3.2 days and 5.7 days
respectively in the computer assisted group compared with 4.5 days and
9.4 days respectively in the control group.11 They used a
different definition of a maintenance dose.
Control of maintenance dose of warfarin
The intervention group spent more time within the therapeutic
range as inpatients and outpatients. Although there was a tendency to
undertreat patients in the control group, there was no increase in the
number of thromboembolic events. However, a much larger study would be
required to show any difference. Also we did not investigate the longer
term effects of undertreatment such as chronic venous insufficiency and
leg ulcers.
Other randomised studies comparing computer assisted dosing with
treatment by doctors or nurse practitioners alone have not reported any
difference in the quality of anticoagulant control.1012
Poller et al found no difference between computer
assisted groups and their control group in the proportion of measured
international normalised ratios being within the therapeutic range
(51.5-59.7%, P=0.62).10 In contrast, we found that, for
outpatients, the percentage of ratios within the therapeutic range was
44% in the control group and 58% in the intervention group.
Limitations of study
The two patient groups in our study were similar in size, clinical
characteristics, and duration of follow up. Other factors, however,
such as change in hazards over time and outpatient compliance with
treatment were not determined satisfactorily and were assumed to be
similar in both groups.
Because of logistical problems, it was difficult to shield the doctors
treating the control group from the computerised decision support
system's suggestions, especially for outpatient control. Hence, there
may have been some learning effect and therefore a carry over effect in
the treatment of the control group. The doctors could have been
randomised to the different groups, but this would have been difficult
as there was a high turnover during the study and the study groups
might have differed in size and hazards. We did, however, control for
factors such as illicit use of the aid by doctors treating the control
group, the problems of unfamiliar technology, and feedback from the
decision support system about previous performance.
| Key messages |
- The quality of control of warfarin doses during initiation and
maintenance of oral anticoagulation is generally poor
- We investigated whether a computerised decision support system
for initiation and control of oral anticoagulation improved quality of
anticoagulant control achieved by trainee doctors
- The median time to achieve a stable dose was significantly
lower in the group assisted by the decision support system than in
controls, without excessive overtreatment or undertreatment with
anticoagulant
- Patients in the group with the decision aid spent more time
within the therapeutic range both as inpatients and outpatients
- The computerised decision support system was safe and
effective and improved quality of initiation and control of warfarin
treatment by trainee doctors
|
- Conclusion
Our study shows that our computerised decision support system was
safe and effective in the initiation and control of warfarin treatment.
It is difficult to predict whether the same use and acceptance of the
system would occur if it were made available directly to trainee
doctors. A fuller evaluation of the system - including its direct
usability, clinical effectiveness, and cost effectiveness - is
necessary. However, it has the potential to improve the quality of
anticoagulant control by trainee doctors as well as allowing dosing and
monitoring in primary care.24
- References
- 1 Duxbury B McD. Therapeutic control of anticoagulant
treatment. BMJ 1982;284:702-4.
2 Majumder G, Payne R W. Quality of oral anticoagulant therapy.
Clin Lab Haematol 1985;7:125-31.
3 Doble N, Baron J H. Anticoagulation control with warfarin by
junior doctors. J R Soc Med 1987;80:627.
4 Tan G, Cohen H, Taylor F, Gabbay J. Audit of start of
anticoagulation treatment in inpatients. J Clin Path
1993;46:67-71.
5 Rose P E. Audit of oral anticoagulant treatment. J Clin
Pathol 1993;46:1069-70.
6 Holford N H G. Clinical pharmacokinetics and pharmacodynamics of
warfarin - Understanding the dose-effect relationship. Clin
Pharmacokinet 1986;11:483-504.
7 Fennerty A, Dolben J, Thomas P, Backhouse G, Bentley D P,
Campbell I A, et al. Flexible induction dose regimen for
warfarin and prediction of maintenance dose. BMJ
1984;288:1268-70.
8 How to anticoagulate. Drug Ther Bull
1992;30:77-80.
9 Doecke C J, Cosh D G, Gallus A S. Standardised initial warfarin
treatment: evaluation of initial treatment response and maintenance
dose prediction by randomised trial, and risk factors for an excessive
warfarin response. Aust N Z J Med 1991;21:319-24.
10 Poller L, Wright D, Rowlands M. Prospective comparative study
of computer programs used for management of warfarin. J Clin
Pathol 1993;46:299-303.
11 White R H, Hong R, Venook A, Daschbach M M, Murray W, Mungall DR,
et al. Initiation of warfarin therapy: comparison of
physician dosing with computer assisted dosing. J Gen Intern
Med 1987;2:141-8.
12 White R H, Mungall D R. Outpatient management of warfarin
therapy: comparison of computer-predicted dose adjustment to skilled
professional care. Ther Drug Monit 1991;13:46-50.
13 Vozeh S, Steimer J L. Feedback control mechanisms for drug
dosage optimisation: concepts, classification and clinical application.
Clin Pharmacokinet 1985;10:457-76.
14 Vadher B, Patterson D L H, Leaning M. Prediction of INR and
maintenance dose during initiation of warfarin therapy using a Bayesian
pharmacokinetic/pharmacodynamic model. Br J Haematol
1996;93(suppl 1):30.
15 Vadher B D, Patterson D L H, Leaning M S. Validation of an
algorithm for oral anticoagulant dosing and appointment scheduling.
Clin Lab Haematol 1995;17:339-45.
16 Altman D. Practical statistics for medical
research. London: Chapman and Hall, 1991.
17 Lewis S M. External quality assessment. In: Lewis S M, Verwilghen
RL, eds. Quality assurance in haematology. London:
Balliére Tindall, 1988:151-75.
18 Department of Health. Anticoagulant treatment.
London: HMSO, 1985.
19 Rosendaal F R, Cannegieter S C, van der Meer F J M, Briet E. A
method to determine the optimal intensity of oral anticoagulant
therapy. Thromb Haemost 1993;69:236-9.
20 Glynn R J, Buring J E. Ways of measuring rates of recurrent
events. BMJ 1996;312:364-7.
21 Stukel T A, Glynn R J, Fisher E S, Sharp S M, Lu-Yao G, Wennberg
J E. Standardised rates of recurrent events. Stat Med
1994;13:1781-91.
22 Abbrecht P H, Timothy J O, Behrendt D M. Evaluation of a
computer-assisted method for individualized anticoagulation:
retrospective and prospective studies with a pharmacodynamic model.
Clin Pharmacol Ther 1982;32:129-36.
23 Carter B L, Taylor J W, Becker A. Evaluation of three
dosage-prediction methods for initial in-hospital stabilization of
warfarin therapy. Clin Pharmacol 1987;6:37-45.
24 Fitzmaurice D A, Hobbs F D R, Murray J A. Monitoring oral
anticoagulation in primary care. BMJ 1996;312:1431-2.
(Accepted 26 March 1997)
We thank Professor S Gallivan, Clinical Operational
Research Unit, University College London, for statistical advice.
Funding: North Thames Regional Health Authority Research and
Development.
Conflict of interest: None.
Cardiovascular Department, Whittington Hospital,
London N19 5NF
B Vadher, research
registrar
D L H Patterson, consultant
cardiologist
Centre for Health Informatics and
Multiprofessional Education (CHIME),
University College London,
Whittington Hospital,
London N19 5NF
M
Leaning, senior lecturer
Correspondence to: Dr
Vadher.
