The study is an alternate time series experiment which consists of three successive six-month periods (timed to change with the changing of medical residents even though they will not be the only prescribers). Each phase will consist of:
• a two-month period without active support during which evaluation criteria will be collected (period A),
• a two-month period with non-personalised active support (period B),
• a two-month period with personalised active support (period C).
To limit the impact of a learning effect on appropriate OAT management practice within the department over time (possible for medical residents), the order of these three periods was determined by using a Latin square plan (see Figure). This experimental design can be considered valid for an impact study in this context whereas a randomised controlled design is difficult to apply with just one hospital [20
This study will include all the patients who are prescribed OAT for any indication and are hospitalised in clinical care departments where physicians are using the hospital information system to prescribe drugs.
The following table shows the number of INR examinations prescribed by these departments in a three-month period. It provides an estimation of the approximate proportion of overdoses among the INR that exceeded 2, which is supposed to be found in patients treated with OAT in these units.
All doctors authorised to prescribe drugs in the participating departments will be included in the study: residents and fellows, registered and non registered university hospital doctors. Each six-month period will coincide with an internship semester.
To prescribe a drug using Dx-Care®, the doctor selects the required drug from an exhaustive list. This opens up a dialogue box in which the doctor types the dose, the frequency of intake and the mode of administration. From this window, it is possible to add a text comment or to consult particular protocols that have been defined by the departments.
It is planned to integrate two types of decision support systems into the computerised prescription program:
1) non-personalised active system: when the drug is selected a window automatically opens giving the prescribing the nomogram for the adjustment of OAT doses in the form of a table (see Tables and ).
2) personalised active system: when the drug is selected a window automatically opens suggesting a dose recommended according to the nomogram (taking into account the doses previously received by the patient and the patient's INR), together with a date for next INR control and an explication.
Definition of endpoints
Proportion of patient observation time with INR results > 4.5, assuming linear change of INRs.
Major haemorrhagic accidents
Intra-cranial haemorrhage or spontaneous haemorrhage necessitating surgery or a transfusion or decreasing haemoglobin concentration by more than 2 g/dl.
Assessment of evaluation criteria
The Netlab® application allows biological laboratories to receive prescriptions and to return results. All of the INR results can be extracted from the Netlab® database accompanied by information making it possible to identify the patient, the treatment and dose received, the prescribing doctor, the hospitalisation unit, the date the test was prescribed. Data about overdoses can therefore be collected systematically by regular database searches.
Furthermore, the storage of the information in a computerised tool will make it possible to determine previous doses and INR results each time a drug is prescribed.
When a health care professional decides to declare an undesirable event, he or she fills in a specific, pre-formatted form available on the Intranet. This form includes a list of events that must be declared at the GPEH.
The declaration form includes an item entitled "complication of haemorrhagic accidents". When the doctor clicks on this item, a specific form for the declaration of a haemorrhagic accident associated with anti-coagulant treatment appears (see form in appendix).
Determination of sample size
The determination of the number of participants necessary requires the definition of the statistical unit of interest, information about the incidence of the evaluation criteria in the study population and a hypothesis about the efficiency of the intervention.
In this study, the main aims are to guide each prescription and to reduce the number of anti-coagulant overdoses: the simplest statistical unit to study is therefore the INR result. This unit will be used to calculate the sample size.
This choice is not, however, perfect and the efficacy results will be presented using other indicators of the quality control of anti-coagulant treatments:
Given the low incidence of major haemorrhagic accidents (not currently measured at the GPEH but probably below 1%), it is not possible in this study to estimate the number of subjects necessary to demonstrate an effect of intervention on the "haemorrhagic accident" endpoint. Recording haemorrhagic accidents will give the frequency of such accidents, which will then be used for realistic estimates of power and sample size if further studies are carried out.
In previous studies evaluating the efficacy of tools to aid the prescription of OAT, the unit considered was not always the same, taking into account the number of INR per patient and the time between INR measurements to greater or lesser extents. The most recent studies considered the number of patient-days according to the method described by Rosendaal [15
]. This method can also be used to calculate the rate of haemorrhagic events as a function of the number of patient-days for a given range of INR values.
We may also carry out an analysis for each prescribing doctor given that the intervention targets doctors directly. This will involve adjusting the effect of the intervention to the fact that intra-physician variability is a priori lower than inter-physician variability.
Number of INR measurements and predicted frequency of overdoses
During a six-month period (January to June 2004), 4 920 INRs were requested by the six departments which already routinely use the computerized physician order entry system. The frequency of overtreatment can approximately be estimated from the percentage of INR > 4.5 among INR >2. Among the 2620 INR > 2, 330 (12%) were higher than 4.5 (see Table ). This frequency has been stable during these six months but differed considerably between departments (10% to 23%).
Results of INR ordered during first semester 2004 in six departments
Hypothesis about the efficacy of the intervention
The number of INR tests necessary for a six-month period, with an α risk of 5% and a power of 80%, for the comparison of two percentages by classical methods (untreated group half the size of the treated group), for a basal incidence of the judgement criterion of 12% are and for the following hypotheses on relative reduction of the risk (RRR) of overdose, are:
• RRR 30%: 2500
• RRR 40%: 1300
• RRR 50%: 800
• RRR 60%: 500
Carrying out approximately 5000 tests over six months will make it possible to detect an intervention effect of less than 30% in this period. The experimental design includes three six-month periods and should thus ensure adequate power.
Statistical analyses will be performed with the STATA statistical software (release 8, STATA corp, College station, Tex, USA)
Standard statistical tests will be used to compare the baseline characteristics of the departments and patients.
The main analysis concerns the effect of the intervention on the number of dangerously high INRs. The analysis will be carried out using a mixed effect analysis of variance model, in which the effect linked to the period will be considered fixed and that linked to the prescription tool will be considered random [22
Rosendaal's method will be used to analyse the number of patient-days with INR over the target [21