times 2 weeks cross over study in 8 individuals with
3 GENERAL METHODS AND SOURCES OF DATA
3.1 Economic Evaluation
3.1.5 Testing for uncertainty
3.1.6.1 Concerns about modelling
Concerns about modelling usually focus on the inappropriate use o f clinical data, the transparency or validity o f the model and the ease w ith which differences between treatment groups can be made to appear statistically significant. By their very nature, models usually consist o f data from more than once source. However, it is important to ask how these heterogeneous pieces o f information were derived because they might not be accurate estimates o f the particular variable in question; a ‘good’ economic evaluation should at least reference each source. For example, Buxton’^'’ notes the problem o f extrapolating from observational data and the role o f potential biases in deriving accurate future projections o f outcomes. There is also some concern over the transparency or validity o f models. Often due to publication space, models are not
always fully described in a publication thus, it is sometimes difficult to conceptualise exactly how a particular question has been examined. The issue o f statistical inference is also an important issue. This is because increasing the number o f hypothetical individuals entering a model increases the sample size and narrows the standard error. Thus, it is conceivable that a model could be run purely until differences appeared to be statistically significant and inaccurate conclusions drawn as a result. However, as Buxton rightly says, sometimes modelling appears to be an unavoidable fact o f life'^"^.
Two o f the economic evaluations presented in this thesis have been performed using modelling techniques because o f the need to combine information from a number o f different sources. Both models are based on decision analytical techniques as it provides an intuitive framework for evaluation^°\ Decision trees are useful in situations where all relevant events happen over a relatively short period o f time; otherwise they can become too large to handle efficiently. Markov models, however, are perhaps more appropriate in situations which involve a continuous risk over time (although values can be time-dependent if specified), when the timing o f events is important or when important events may occur more than once^^^'^^"^. The disadvantage o f this approach is, however, that Markov models do not have ‘mem ory’ meaning that the probability o f moving to a health state in the following cycle is purely dependent on the probability assigned to the current health state; the so-called chain rule.
Both types o f modelling combine data on possible outcomes (costs and benefits) with the probability o f relevant clinical and economic events occurring to produce expected values. For example, if the National Lottery paid out 25% o f all wagers in cash prizes, every £1 spent on Lottery tickets has an expected return o f £0.25. This does not mean that a holder o f a £1 Lottery ticket you will always get a return o f £0.25, but that if he / she plays the Lottery many times, there will be m average return o f £0.25 per £1 purchased ticket.
the necessary clinical issue(s). The paragraphs below describe the sources o f these data. However, more detailed descriptions o f the additional collected data and the methods used to collect them are presented in the relevant empirical chapters.
3.2.1 The Royal Free Hampstead NHS Trust
The Katharine Dormandy Haemophilia Centre (KDHC) was established at the Royal Free Hampstead NHS Trust in 1964 for five outpatients under the direction o f Dr. Katharine Dormandy. Since this time the centre has seen enormous expansion and now provides treatment for over 1,500 individuals with various congenital clotting factor disorders. The KDHC is one o f 26 comprehensive care centres (CCCs) in the UK. To qualify as a CCC, treatment must be provided for 40 or more severely affected individuals with haemophilia per year and access must be provided to other specialist services eg. orthopaedic units, HIV and hepatitis expertise, counselling and physiotherapy\ Individuals attend from a variety o f geographic areas, including some from outside o f the UK, and treatment is provided for patients o f all ages, including children. Since the early 1990s it has been policy to place, whenever feasible, all previously untreated patients (PUPs) with severe haemophilia on primary prophylaxis.
In 1980, under the direction o f Dr Peter Kemoff, an internal Paradox (Borland software) database was installed at the KDHC. Data that have been recorded on the database since this time include the type and amount o f clotting factor concentrate administered, the manufacturer, the reason for treatment (eg. for surgery, prophylaxis or following a bleed), the date treatment was administered and the details o f bleeds. Specific details o f each bleed include the date o f each subsequent clotting factor dose, the number o f doses o f clotting factor required to treat each bleed and the size (in iu) o f each dose, the type o f clotting factor used and the location o f each bleed. Thus, the number o f bleeds an individual expereienced could be calculated by summing the number o f ‘first’ infusions where treatment on-demand was also indicated. Instances where no bleeds were recorded on the database for a particular individual were assumed to indicate that they had not experienced any bleeds over that period o f time (ie. from the date o f registration at the KDHC to the end o f the period under investigation for the particular analysis).
Individuals are required to record details o f all infusions o f clotting factor administered outside o f the hospital onto a specially designed ‘home-infusion’ record. Patients return
the database. However, the database was not primarily designed with research in mind, thus initially, considerable time was spent arranging the information on the database into a format that could be imputed into a statistical package.
Data items from this database that have been analysed in this thesis are shown in Table 3.1. Further data related to individuals registered at the KDHC that were also analysed in this thesis were collected directly from the patients’ medical notes and from the Royal Free Hampstead NHS Trusts’ Patient Admissions System (PAS). The PAS system was used to collect retrospective data on hospital visits for the period 1988 to 1997 inclusive. More specifically, details on the date o f attendance, the treatment speciality (eg. haemophilia, orthopaedic, dental etc.), the type o f attendance (inpatient, outpatient or day case) and the length o f inpatient stays were collected. However, as with the information from the KDHCs internal database, considerable time was spent collecting and manipulating these data into a useful format as individual patient based data could be down-loaded from the PAS system and because patient records can only be searched one at a time.
The remaining data on Health-Related Quality-of-Life (HR-QoL) and the patient / family costs associated with haemophilia were collected cross-sectionally by posting appropriate questionnaires to selected sub-groups o f individuals who were registered for treatment at that KDHC. More specific details regarding these data and the methods used to collect them are provided in the appropriate empirical chapters.
Table 3.1: Data collected and analysed from individuals registered at the KDHC
Data type
Variable
Source
Demographic details Haemophilia type / severity
Weight (over time) Date o f birth
Database Medical notes Medical notes
Health-related Quality-of-life Postal survey
Inpatient stays /date / reason / length PAS
Inform ation from the Katharine Dormandy Haemophilia Centres internal database covered the period 1980 to 1997 inclusive
* PAS - Patients Admissions System
3.2.2 Other data sources used in this thesis