Journal of Medical Engineering & Technology
ISSN: 0309-1902 (Print) 1464-522X (Online) Journal homepage: http://www.tandfonline.com/loi/ijmt20
Valve data collection: problems and pitfalls P. J. Drury, M. M. Black, C. J. Ashman & J. Piercey To cite this article: P. J. Drury, M. M. Black, C. J. Ashman & J. Piercey (1992) Valve data collection: problems and pitfalls, Journal of Medical Engineering & Technology, 16:1, 4-9, DOI: 10.3109/03091909209021950 To link to this article: http://dx.doi.org/10.3109/03091909209021950
Published online: 09 Jul 2009.
Submit your article to this journal
Article views: 1
View related articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ijmt20 Download by: [Universität Osnabrueck]
Date: 14 March 2016, At: 21:59
Journal of Medical Engineering & Technology, Volume 16, Number 1 (January/February 1992), pages 4-9
Valve data collection: problems and pitfalls P. J. Drury, M. M. Black?, C. J. Ashman$, and J. Piercey*
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Institute f o r Biomedical Equipment Evaluation and Services, Lodge Moor Hospital, Shefield SIO 2LH, UK ?Department of Medical Physics and Clinical Engineering, Lhioersity .f SheJfield, Shefield SIO 2JF, UK $Statistical Services Unit, University of Shefield, ShefJicld S3 7RH, LIK *Centre for Health Economics, University of York, York YO1 5 0 0 , LIK
Since 1981, the Department of Medical Physics and Clinical Engineering at the University of Shefjfild has been responsible f o r the organization, management and data collection associated with the largest multicentre heart valve implant patient followup study in the Western world. A t the present time, the database comprises information on over 16,000 valve implants, which have been provided by 57 surgeons working at 22 centres in the UK. All this data is available f o r in-depth statistical analysis. Over 30 individual valve models presently are included in the Study and these can be categorized into Jive main types: ball, disc, porcine, pericardial and homograft. Analysis includes descriptive statistics as well as valuable information on the various peformances of the different valves. Survival and event-free survival graphs are obtained by actuarial methods and individual valve types can be studied in depth in terms of freedom from thromboembolic complications and valve dysfunction. Whilst this approach provides interesting and valuable survival data, it does not take account of the wide variation in prognostic factors which occur within large groups of patients. This latter problem can be addressed by the use of proportional hazards analysis and this paper provides details of this approach and typical results obtained from the use of this method. These include !he comparative perfOrmances of the major Qpes of valves currently in use in terms of the event-jree survival of the patients.
Introduction In vitro test facilities are widely available for assessing the performance of prosthetic valves, but the data they provide cannot be used to predict corresponding in vivo performance. The only reliable way of obtaining this information is by long-term follow-up of valve implant patients. Large amounts of data are needed and individual centres may require many years of data collection in order to build up an adequate databank. Alternatively, a large number of centres can participate in a joint approach to such data collection, with each centre using identical proformas for data input. After detailed discussions during the early 1970s, it was decided that the most appropriate approach to the problem in the UK would be to set up a multicentre valve replacement study to collate valve implant and follow-up data. Such a study, the U K Multicentre Valve Study, was initiated in 1974 and is based in the Department of Medical Physics and Clinical Engineering at the Royal Hallamshire Hospita!, Sheffield, UK.
The Multicentre Valve Study, now in its 18th year, is processing data on 16,359 valve implants in 13,852 patients. These data have been received from 22 centres involving a total of 57 surgeons. Follow-up details are available on 12,644 valves. This figure represents 77.3% of total valve numbers. There is a maximum follow-up of 23.8 years with a mean of 4.2 years for individual valves and a mean of 4.3 years for individual patients (responders only). The data have established the project as being unique in terms of its patient numbers and range of valves. It is still the largest combined study of its kind and deals with patients from many different parts of the country and, in that sense, it is reasonably representative of heart valve implantation in the UK.
Organization of the Study Organization and day-to-day running of the Multicentre Valve Study are arranged to provide a personal and confidential data service for the participating surgeons, to analyse and publish the pooled data on a regular basis and to offer an information service using the pooled data store. Another objective of this study is to follow closely the performances of new valve models. Including a large number of centres provides more detailed and accurate reporting of early follow-up results for these valves rather than a series of often conflicting reports from many different centres. In order to collect data, a number of forms arc uscd:
Patient discharge summary. This is completed by the surgeon, provided the patient is discharged from hospital. Data provided on this form gives the patient’s name, age and sex, hospital and surgeon, preoperative status of the patient and valve replacement operative data. Follow-up summaries. These are completed on successive anniversaries of the valve replacement operation and require the use of one or more of the following forms: No event. This form is completed provided the patient has had a complication-free year and records details of the status of the patient including any anticoagulation therapy. Thromboembolism, embolus and haemorrhage. This form is completed if one or more of these events occurred during the previous year. Details of the site of the event, its effect on the patient, any action taken subsequently and the present status of the patient are recorded.
P.J. Drury
el al. Valve data collection: problems and pitfalls
Valve dysfunction. This form is completed if the surgeon considered the valve substitute to havc failed to function adequately. Causes of dysfunction include infection, leakage around the implant, thrombus formation, dislodgement or wear of the valve or its components, and disruption of any part of the implant. Effects of such incidents on the patient, any action taken subsequently to resolve the condition and the present status of the patient are recorded.
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Patient mortality. This form is completed on the death of the patient. Details ofdisease or condition directly leading to death and any other significant conditions contributing to dcath, but not related to the cardiac condition of the patient, are recorded.
Mitral 7017 (42.9%)
Aortic 8892 (54.4%)
Figure 2a. Proportion of implanted valves in each valve position.
Results of data analysis As already noted, the Multicentre Valve Study processes data on many thousands of valve implant patients. Age distribution of these patients, both in total and by sex, is given (figure la), while figure Ib indicates the variation with age of the ratio of aortic-to-mitral replacements for male and female patients.
Of the implants, 54.4% have been in the aortic position, 42.9% in the mitral, 1.5% in the tricuspid and 1.2% in the pulmonary (figure 2a). Figure 2b shows the most commonly used valves in the period during which data have been collected. The majority of valves have been either porcine (36.1YO) or single-leaflet disc valves (33.0%), with pericardial (8.3%) and ball valves (10.1%); and double-leaflet disc valves (6.3%) and hornografts( 5.3%) being used in comparable proportions
Valve Type
Figure 2b. Most popular valves.
irm, 12m
Mo
OMale
7184
Figure 2c. Proportion o f each lrpe o f valve used. 400
m 100
0 D5 b10 11-15 15.m 21.25 26.30 31-35 36-4041-454b50 51-55 56-M 6165 66-7071-75 7bBO 8lsS S-90
Figure l a . Age distribution by sex of patients at Jirst surgery. % of valve implents
-
-0.2u -0.44 -
4
Figure 2d. Number of each valve p p e used each year. -V
0 . 10 is P B J ) 3s Q Q
6
so ISUI
1 6 1015 80
Patient age at surgery
Figure 16. Ratio of aortdmilral valve implants by age and sex.
(figure 2c). The trend in valve usage is independent of follow-up period. Distribution of the major valve types by year of entry into the Study is given (figure 2d) and illustrates clearly the rise in the number of bioprosthetic valve implants from 1974 to 1981. In 1981, biopros5
P. J. Drury el al. Valve data collection: problems and pitfalls
thetic valves accounted for 65% of all implants being recorded by the Multicentre Valve Study. Since that time there has been a gradual decline in the number of bioprostheses implanted, with a concomitant rise in the use of mechanical valves.
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Many different aspects of valve substitute implantation have been quantitatively assessed from the data recorded by the Multicentre Valve Study [l-121 and one of the major objectives of this study is the provision of comparative performance data on the various valves that predominate in routine clinical use. One of the ways in which such information may be obtained is by the use of the statistical method of actuarial analysis. Actuarial curves of probability of patient and eventfree survival after aortic, mitral and multiple-valve replacement are given (figure 3). Actuarial curves of freedom from valve dysfunction (figure 4) and a comparison of the incidences of embolic, haemorrhage and overall thromboembolic events, irrespective of implant position (figure 5) are also given.
'I
20
I 2
1
n
. o
(1322)
(8W
(508)
(350)
(172)
(01)
(lib
(im)
(ioia)
(55q
(zu)
(72)
(1s
(0)
la921
(252)
~212)
(101)
LlW)
(6.3)
(25)
2
4
8
10
12
14
0
6
1 I ~
16
Follow-up (years)
Figure 4a. Probability of freedom from valve dysfunction a)er aortic valve replacement with mechanical (0)) bioprosthetic (0) and free homograft valves (.).
100
P
: 80 c
60 c
40
0
2:
P
=
Pis g
20
01 0
0
(1182)
(680)
2
4
(356)
(143)
(42)
(0)
(3)
(2)
8
10
12
14
16
I
6
Follow-up (years)
Figure 46. Probability of freedom from valve dysfunction afier mitral valve replacement with mechanical (0)and bioprosthetic (0) valves.
' . 6 0
60-
i?
I
PW
(887)
Wl
o (2593) (lW5) (10%)
(040)
(355) (102)
(15w ~1WO) (010)
(3%)
(1W
(2339) (1490)
1181)
(En)
(63)
(21)
17)
(43)
(15)
(2)
(19)
(0
100
g
80
a
70
P
0
2
4
6
8
10
12
14
16
18
20
Follow-up (years)
.B n
01 0
Figure 5 a . Probability offreedom from embolic (a))haemorrhage (0) and thromboembolic (I)events after valve replacement with mechanical valves.
.
(1475)
w)
(son)
2
4
6
(no)
8
1124)
(40)
(lo)
10 12 14 Follow-up (years)
(0)
16
~
18
Figure 36. Probaility of event-jree survival after aortic mitral (0) and multiple (.) valve refilacement. 6
j
20
Statistical assessment of follow-up data
(a),
As already discussed, valve performance is usually evaluated by actuarial analysis. Comparisons of the performance of valves are then achieved by visual
P. J. Drury
P
el al. Valve data collection: problems and pitfalls
o (3557)
50
(?la) (12M)
0 ~ 7 1(~017) ( i w
0
2
4
6
(572)
(2m)
(41)
(17)
(11)
(1)
(515)
(119)
(57)
(141
(7)
(1)
8
10
12
14
16
18
’
20
Follow-up (years)
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Figure 56. Probability OJ‘freedomfrom embolic (e),haemorrhage (0) and thromboembolic (m) events afier valve replacement with bioprosthetir valves.
comparisons of thc actuarial curves for the different valves. Although this method does provide a guide to valve performance, such as event-free survival, valve dysfunction or mortality, it has certain limitations. First, it is not possible to account for differences in underlying patient prognosis. For example, if those patients receiving a certain model of valve have a poor prognosis compared with those receiving a different modcl of valve, the observed differences betwcen the actuarial curves for the two valves could bc due to differences between the valve models or to diffeercnces in patients prognosis. Second, even when standard crrors are attached to the actuarial estimates, it is difficult to make valid comparisons betwcen them that allow statements to be made about thc statistical significance of their difference. Use of proportional hazards modelling, however, permits a formal statistical comparison of different valves. This is a model that allows for the Significance of different prognostic factors to be considered. Each of these various factors can be looked at in turn so one can determine which onc accounts for the most variability in the response time. Thereafter, additional factors are included in the model to sce which, if any, improvcs it. This procedure is followed in a stcpwise fashion with all thc individual factors until a stage is reachcd when no further improvement to the model is possible. In this way, it is possible to use a large databank comprising many data on prognostic factors to determine which are the most important factors in relation to, for example, valve dysfunction, cvent-free survival or mortality. This analysis cannot bc achicvcd by the use of the life table/actuarial approach. Proportional hazards modelling has been used on Study data in an attempt to dctermine which preoperative factors could be rclatcd with time to valve dysfunction (valve dysfunction being cither valve failurc or infection, thromboembolism or haemorrhagc). As alrcady noted, the advantagcs of this typc of analysis over other methods arc that it allows the assessment of the effect of several preoperative factors acting simultaneously and provides an estimate of an individual’s projected time
to valve dysfunction given a composite profile of their preoperative characteristics. Figure 6 shows the estimated survival times for one set of prognostic factors. Each factor has a specific and significant effect on time to valve dysfunction. Many other similar plots can be produced by varying the pattern of these prognostic factors. For example, it may be more desirable to determine estimated survival times for patients with the same prognostic pattern, as shown in figure 6, except that they are aged over, not under 56 years. Figure 6 does, in fact, represent a slightly complicated model because rather than one survival function for a given set of prognostic factors there are five. T h e reason for this ‘stratified’ model is because the prognostic factor ‘valve type’ violated the assumption of proportionality. Proportionality means that any differences between the individual levels within a prognostic factor remain constant over time. Clearly, this is not the case here since the survival functions of disc, pig and tissue valves cross between 60 and 80 months. There are many possible subsets of the data which could be analysed using the proportional hazards model. Owing to the size of the Study such analyses are possible with little loss in power of the statistical tests employed. However, results of any such study using non-random data should always be treated tentatively. As with all statistical analysis based on the use on non-random data, care must always be taken in any interpretation involving behavioural prediction.
1 .o 0.9
z
t
0.8
U
5
6
f
0.7
0.6
E
8 81 f
0.5
c
O
&
0.4
a0
ng n
0.3
-A = -.- B C = --D = .----. E
=
Ball (558) Disc(2794) Pig (3026) lissue(747) Homograft(208)
\
\ \ \ \ \ \
\ \
\ \
\
A
\
\ Prognosticpattern D Age = Under 56 No previous embolism Not aortic stenotic No other operative diagnosis Aortic implant NYHC = 1
0.2 0.1
I
I
1
20
40
60
1
,
I
I
1
1
80 100 120 140 160 180 200
FOIIOW-up (months)
Figure 6. Proportional hazards analysis - estimated survival function. 7
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
P. J. D r u 7
ct
al. Valve data collection: problems and pitfalls
Comparison of cost-effectiveness of different types of heart valve
operation, cost of any ongoing therapy and cost of any subsequent valve-related event. The main benefit is a measure of survival after discharge.
The Study, as already noted, contains a wealth of data about almost 14,000 patients who have undergone heart valve replacement. Of particular importance is the availability of data concerning valve performance. Clinical performance of heart valves has been assessed fairly comprehensively but, to the best of the authors’ knowledge, no attempt has yet been made to perform a cost-effectiveness analysis of this field. Another aim of the Study, therefore, is to compare the relative costeffectiveness of using different types of heart valve. Though there are several types of heart valve, and within each type there are a variety of specifications and makes, this initial investigation focuses on a comparison between the two most commonly used types; mechanical and bioprosthetic valves. Later analyses will include different types and makes of valve.
Even though the valves can be considered to be direct substitutes in the vast majority of cases, their economic consequences are different (substantially). Mechanical valves, once fitted, can last for 20 years or more before re-replacement, whereas bioprosthetic valves last around 7 years. However, once fitted, bioprosthetic valves require no ongoing therapy whereas mechanical valves require some drug therapy, and so have a higher ongoing cost which counteracts the cost of more frequent replacement of the bioprosthetic valves. Figure 7 shows the two cost profiles (at current prices) over time. Both types of valve can be the cause of medical problems; events like thromboembolism and haemorrhage, infection and valve failure, and sometimes death. The Study has a record of such postoperative events and complications. These complications need to be treated and in all probability will require an inpatient stay. A significant cost will then be borne by the NHS. This cost must taken into account when comparing the costeffectiveness of the two types of valve. In order to estimate the additional cost imposed by each of these events, information is required about the likelihood of these events occurring. Using the database, the probability of any particular event occurring in any particular year after the initial operation can be estimated. Cost of the event in current prices, together with the probability of the event occurring, allows the expected cost of events in any given year to be included in thc cost profile. Thus, a cost profile (in current prices) can be built up to include operating costs, ongoing therapy and adverse events. However, costs relating to the two valve types are incurred at different times. Given that costs incurred in the future are worth less than costs incurred in the present, because of time preferences, some form of discounting needs to be introduced. Using the appropriate method of discounting enables expected lifetime costs to be built up for each valve type. These expected lifetime costs can be expressed in current prices, and thus are comparable directly.
The most appropriate method of carrying out an evaluation of this type is cost-effectiveness analysis. In a cost-effectiveness analysis, both costs and consequences of treatment in question are examined. It differs from simple cost or cost-minimization analyses in that it does not assume that the outcome differences of the two treatments are identical. O n the other hand, it does not require an explicit valuation to be made of the differing outcomes (if there are any differences) which a costutility analysis would require. A cost-effectiveness analysis allows the differences to be expressed in natural units (for example, life years saved) whereas a costutility analysis would require a valuation of those life years. Hence, no problem arises concerning assigning different valuations of life years to different groups of people. The two differing and most common forms of treatment that will be under consideration are valve replacements using mechanical and bioprosthetic valves respectively. It is proposed that homografts and other options will be included in a later study. Some patients may be treated with drug therapy as opposed to implantation, but this is a short-run option and these patients tend to be elderly or unsuitable for valve replacement. Generally, there is no ‘do nothing’ alternative; prospective patients cannot survive as long or with as good quality of life without a valve replacement. In addition, mechanical and bioprosthetic valves can be regarded as genuine options. There are only two significant exceptions. The first concerns patients who cannot tolerate anti-coagulant therapy. These patients will not be implanted with a mechanical valve, as ongoing anticoagulant therapy is required. O n the other hand, teenage and younger patients rarely receive bioprosthetic valves because of the higher risk of re-calcification. These two groups only account for a small number of patients though, for over 90% of operations there is a free choice between mechanical and bioprosthetic valves.
Cost per year
0
Given that a cost-effectiveness approach has been chosen, the next step is to identify the relevant cost and effectiveness data. These include the cost of the initial 8
5
Current prices
10
15
20
25
Year
Figure 7. Cost profiles of mechanical and bioprosthetic values.
P.J. Dmry rt
al. Valve data collrction: problems and pitfalls
Turning to effectiveness, it was established earlier that
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
there was no acceptable ‘do nothing’alternative. Suitable patients are always implanted. Therefore, it becomes possible to estimate the number of life years gained by thinking in terms of survival after discharge. I n addition to simply thinking in terms of life years gained, there is also a crude facility within the database for measuring quality of life; the New York Heart Association classification, which is a method of measuring dysfunction. This can be used to build up a profile of quality of life as provided by dysfunction and it is hoped that this can be used in conjunction with the other data in the assessment of cost-effectiveness. Again, direct comparisons can be made beween the two types of valve. Thus, the Study database contains enough information for a crude assessment of cost-effectiveness. However, some assumptions have to be made, in particular the costs which will need to be obtained, the expected probability of adverse events and the appropriate rate of discount. A comprehensive sensitivity analysis is required to test the effects of varying these assumptions. This method has been designed for a basic study of mechanical and bioprosthetic valves. Methodology is very general and can be easily developed for use in other cardiac procedures. In terms of heart valves, it is envisaged that the work would be extended to include firstly homograft valves and then analysis of different makes of valve within general valve types. Methodology can be also refined further, for example, to include any alterations in the probability of additional events after a first event has taken place and to take account of age and sex in the selection of the most appropriate valve type. What has been presented in this section is a general framework for cost-effectiveness analysis applied to the comparison of different heart valves and utilizing existing data sources. This framework can be readily adapted for use in other fields and expanded to a cost-utility approach if the additional data required becomes readily available.
References 1 . BLACK,M. M., COCHRANE, T., DRURY,P. J. A N D LAWFORD, P. V. (1990) In vitro and in vivo performance of artificial heart valves Proceedings of the Conference on Medtcal and Biological Implant Technology, London, 27-29
March 1990. UK Liaison Committee for Services Allied to Medicine and Biology. 2. BLACK, M. M., COCHRANE, T., DRURY P. J. and LAWFORD P. V. (1987a) Artificial heart valves: past Performance and future prospects Cardiovascular Reviews and Reports. 8 , 40-45. 3. BLACK, M. M., COCHRANE, T., DRURY, P. J . and LAWFORD P. V. (1987b) Assessing the performance and safety of artificial heart valves. Proceedings of the 9th E M B S Conference, 3, 1183-1 184. 4. BLACK,M. M., DRURYP. J. and LAWFORD P. V. (1991) Measurement of the in vitro and in vivo performance of artificial heart valves. In A Concise Enqclopaedia of Biomedical and Biological Measurement Systems ( Pergamon Press, Oxford) p. 187-195. 5. BLACK,M. M., DRURYP. J. and SMITHG . H. (1983) Long-term assessment of heart valve substitutes. L f e Support System, 1 (Suppl. I ) , 301-304. 6. BLACK, M. M., DRURYP. J. and TINDALE, W. B. (1985) The clinical performance of bioprosthetic heart valves. In Biocompatibility of Tissue Analogs. Vol. 2 Edited by D. F. Williams (CRC Press, Boca Raton), pp 173-186. 7. BLACK,M. M., DRURY,P. J. and TINDALE, W. B. (1983) Twenty-five years of heart valve substitutes: a review Journal of the Royal Society of Medicine 76, 667-680. 8. DRURY, P. J., BLACK,M. M., LAWFORD, P.V. and KAY, R. (1987) The long-term clinical assessment of heart valve substitutes Engineering in Medicine, 16, 87-94. P. V. and BLACK,M. 9. DRURY, P. J., KAY,R., LAWFORD, M. (1986) Statistical reappraisal of the analysis of heart valve patient follow-up data-the estimation of valve failure rates LiJe Support Systems, 4, (Suppl. I ) , 121-123. 10. FESSATIDIS, I.TH, VASSILIADIS, K. E., MONRO, J. L., Ross, J. K., SHORE,D. F. and DRURY,P. J. (1989) Thirteen years’ evaluation of the Bjork-Shiley isolated mitral valve prosthesis. The Wessex experience. Journal of Cardiovascular Surgery, 30, 957-965. I I . WAIN,W. H. (1991) A unique porcine bioprosthesisthe Wessex experience. Revue Europeenne de Technologie Biomedicale, 13, 1 1 3 - 1 18. 12. WALESBY, R. (1983) A surgical assessment of the StarrEdwards mitral prosthesis. Current Medical Literature: Cardiovascular Medicine, 2, 65-67.
9
ISSN: 0309-1902 (Print) 1464-522X (Online) Journal homepage: http://www.tandfonline.com/loi/ijmt20
Valve data collection: problems and pitfalls P. J. Drury, M. M. Black, C. J. Ashman & J. Piercey To cite this article: P. J. Drury, M. M. Black, C. J. Ashman & J. Piercey (1992) Valve data collection: problems and pitfalls, Journal of Medical Engineering & Technology, 16:1, 4-9, DOI: 10.3109/03091909209021950 To link to this article: http://dx.doi.org/10.3109/03091909209021950
Published online: 09 Jul 2009.
Submit your article to this journal
Article views: 1
View related articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=ijmt20 Download by: [Universität Osnabrueck]
Date: 14 March 2016, At: 21:59
Journal of Medical Engineering & Technology, Volume 16, Number 1 (January/February 1992), pages 4-9
Valve data collection: problems and pitfalls P. J. Drury, M. M. Black?, C. J. Ashman$, and J. Piercey*
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Institute f o r Biomedical Equipment Evaluation and Services, Lodge Moor Hospital, Shefield SIO 2LH, UK ?Department of Medical Physics and Clinical Engineering, Lhioersity .f SheJfield, Shefield SIO 2JF, UK $Statistical Services Unit, University of Shefield, ShefJicld S3 7RH, LIK *Centre for Health Economics, University of York, York YO1 5 0 0 , LIK
Since 1981, the Department of Medical Physics and Clinical Engineering at the University of Shefjfild has been responsible f o r the organization, management and data collection associated with the largest multicentre heart valve implant patient followup study in the Western world. A t the present time, the database comprises information on over 16,000 valve implants, which have been provided by 57 surgeons working at 22 centres in the UK. All this data is available f o r in-depth statistical analysis. Over 30 individual valve models presently are included in the Study and these can be categorized into Jive main types: ball, disc, porcine, pericardial and homograft. Analysis includes descriptive statistics as well as valuable information on the various peformances of the different valves. Survival and event-free survival graphs are obtained by actuarial methods and individual valve types can be studied in depth in terms of freedom from thromboembolic complications and valve dysfunction. Whilst this approach provides interesting and valuable survival data, it does not take account of the wide variation in prognostic factors which occur within large groups of patients. This latter problem can be addressed by the use of proportional hazards analysis and this paper provides details of this approach and typical results obtained from the use of this method. These include !he comparative perfOrmances of the major Qpes of valves currently in use in terms of the event-jree survival of the patients.
Introduction In vitro test facilities are widely available for assessing the performance of prosthetic valves, but the data they provide cannot be used to predict corresponding in vivo performance. The only reliable way of obtaining this information is by long-term follow-up of valve implant patients. Large amounts of data are needed and individual centres may require many years of data collection in order to build up an adequate databank. Alternatively, a large number of centres can participate in a joint approach to such data collection, with each centre using identical proformas for data input. After detailed discussions during the early 1970s, it was decided that the most appropriate approach to the problem in the UK would be to set up a multicentre valve replacement study to collate valve implant and follow-up data. Such a study, the U K Multicentre Valve Study, was initiated in 1974 and is based in the Department of Medical Physics and Clinical Engineering at the Royal Hallamshire Hospita!, Sheffield, UK.
The Multicentre Valve Study, now in its 18th year, is processing data on 16,359 valve implants in 13,852 patients. These data have been received from 22 centres involving a total of 57 surgeons. Follow-up details are available on 12,644 valves. This figure represents 77.3% of total valve numbers. There is a maximum follow-up of 23.8 years with a mean of 4.2 years for individual valves and a mean of 4.3 years for individual patients (responders only). The data have established the project as being unique in terms of its patient numbers and range of valves. It is still the largest combined study of its kind and deals with patients from many different parts of the country and, in that sense, it is reasonably representative of heart valve implantation in the UK.
Organization of the Study Organization and day-to-day running of the Multicentre Valve Study are arranged to provide a personal and confidential data service for the participating surgeons, to analyse and publish the pooled data on a regular basis and to offer an information service using the pooled data store. Another objective of this study is to follow closely the performances of new valve models. Including a large number of centres provides more detailed and accurate reporting of early follow-up results for these valves rather than a series of often conflicting reports from many different centres. In order to collect data, a number of forms arc uscd:
Patient discharge summary. This is completed by the surgeon, provided the patient is discharged from hospital. Data provided on this form gives the patient’s name, age and sex, hospital and surgeon, preoperative status of the patient and valve replacement operative data. Follow-up summaries. These are completed on successive anniversaries of the valve replacement operation and require the use of one or more of the following forms: No event. This form is completed provided the patient has had a complication-free year and records details of the status of the patient including any anticoagulation therapy. Thromboembolism, embolus and haemorrhage. This form is completed if one or more of these events occurred during the previous year. Details of the site of the event, its effect on the patient, any action taken subsequently and the present status of the patient are recorded.
P.J. Drury
el al. Valve data collection: problems and pitfalls
Valve dysfunction. This form is completed if the surgeon considered the valve substitute to havc failed to function adequately. Causes of dysfunction include infection, leakage around the implant, thrombus formation, dislodgement or wear of the valve or its components, and disruption of any part of the implant. Effects of such incidents on the patient, any action taken subsequently to resolve the condition and the present status of the patient are recorded.
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Patient mortality. This form is completed on the death of the patient. Details ofdisease or condition directly leading to death and any other significant conditions contributing to dcath, but not related to the cardiac condition of the patient, are recorded.
Mitral 7017 (42.9%)
Aortic 8892 (54.4%)
Figure 2a. Proportion of implanted valves in each valve position.
Results of data analysis As already noted, the Multicentre Valve Study processes data on many thousands of valve implant patients. Age distribution of these patients, both in total and by sex, is given (figure la), while figure Ib indicates the variation with age of the ratio of aortic-to-mitral replacements for male and female patients.
Of the implants, 54.4% have been in the aortic position, 42.9% in the mitral, 1.5% in the tricuspid and 1.2% in the pulmonary (figure 2a). Figure 2b shows the most commonly used valves in the period during which data have been collected. The majority of valves have been either porcine (36.1YO) or single-leaflet disc valves (33.0%), with pericardial (8.3%) and ball valves (10.1%); and double-leaflet disc valves (6.3%) and hornografts( 5.3%) being used in comparable proportions
Valve Type
Figure 2b. Most popular valves.
irm, 12m
Mo
OMale
7184
Figure 2c. Proportion o f each lrpe o f valve used. 400
m 100
0 D5 b10 11-15 15.m 21.25 26.30 31-35 36-4041-454b50 51-55 56-M 6165 66-7071-75 7bBO 8lsS S-90
Figure l a . Age distribution by sex of patients at Jirst surgery. % of valve implents
-
-0.2u -0.44 -
4
Figure 2d. Number of each valve p p e used each year. -V
0 . 10 is P B J ) 3s Q Q
6
so ISUI
1 6 1015 80
Patient age at surgery
Figure 16. Ratio of aortdmilral valve implants by age and sex.
(figure 2c). The trend in valve usage is independent of follow-up period. Distribution of the major valve types by year of entry into the Study is given (figure 2d) and illustrates clearly the rise in the number of bioprosthetic valve implants from 1974 to 1981. In 1981, biopros5
P. J. Drury el al. Valve data collection: problems and pitfalls
thetic valves accounted for 65% of all implants being recorded by the Multicentre Valve Study. Since that time there has been a gradual decline in the number of bioprostheses implanted, with a concomitant rise in the use of mechanical valves.
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Many different aspects of valve substitute implantation have been quantitatively assessed from the data recorded by the Multicentre Valve Study [l-121 and one of the major objectives of this study is the provision of comparative performance data on the various valves that predominate in routine clinical use. One of the ways in which such information may be obtained is by the use of the statistical method of actuarial analysis. Actuarial curves of probability of patient and eventfree survival after aortic, mitral and multiple-valve replacement are given (figure 3). Actuarial curves of freedom from valve dysfunction (figure 4) and a comparison of the incidences of embolic, haemorrhage and overall thromboembolic events, irrespective of implant position (figure 5) are also given.
'I
20
I 2
1
n
. o
(1322)
(8W
(508)
(350)
(172)
(01)
(lib
(im)
(ioia)
(55q
(zu)
(72)
(1s
(0)
la921
(252)
~212)
(101)
LlW)
(6.3)
(25)
2
4
8
10
12
14
0
6
1 I ~
16
Follow-up (years)
Figure 4a. Probability of freedom from valve dysfunction a)er aortic valve replacement with mechanical (0)) bioprosthetic (0) and free homograft valves (.).
100
P
: 80 c
60 c
40
0
2:
P
=
Pis g
20
01 0
0
(1182)
(680)
2
4
(356)
(143)
(42)
(0)
(3)
(2)
8
10
12
14
16
I
6
Follow-up (years)
Figure 46. Probability of freedom from valve dysfunction afier mitral valve replacement with mechanical (0)and bioprosthetic (0) valves.
' . 6 0
60-
i?
I
PW
(887)
Wl
o (2593) (lW5) (10%)
(040)
(355) (102)
(15w ~1WO) (010)
(3%)
(1W
(2339) (1490)
1181)
(En)
(63)
(21)
17)
(43)
(15)
(2)
(19)
(0
100
g
80
a
70
P
0
2
4
6
8
10
12
14
16
18
20
Follow-up (years)
.B n
01 0
Figure 5 a . Probability offreedom from embolic (a))haemorrhage (0) and thromboembolic (I)events after valve replacement with mechanical valves.
.
(1475)
w)
(son)
2
4
6
(no)
8
1124)
(40)
(lo)
10 12 14 Follow-up (years)
(0)
16
~
18
Figure 36. Probaility of event-jree survival after aortic mitral (0) and multiple (.) valve refilacement. 6
j
20
Statistical assessment of follow-up data
(a),
As already discussed, valve performance is usually evaluated by actuarial analysis. Comparisons of the performance of valves are then achieved by visual
P. J. Drury
P
el al. Valve data collection: problems and pitfalls
o (3557)
50
(?la) (12M)
0 ~ 7 1(~017) ( i w
0
2
4
6
(572)
(2m)
(41)
(17)
(11)
(1)
(515)
(119)
(57)
(141
(7)
(1)
8
10
12
14
16
18
’
20
Follow-up (years)
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
Figure 56. Probability OJ‘freedomfrom embolic (e),haemorrhage (0) and thromboembolic (m) events afier valve replacement with bioprosthetir valves.
comparisons of thc actuarial curves for the different valves. Although this method does provide a guide to valve performance, such as event-free survival, valve dysfunction or mortality, it has certain limitations. First, it is not possible to account for differences in underlying patient prognosis. For example, if those patients receiving a certain model of valve have a poor prognosis compared with those receiving a different modcl of valve, the observed differences betwcen the actuarial curves for the two valves could bc due to differences between the valve models or to diffeercnces in patients prognosis. Second, even when standard crrors are attached to the actuarial estimates, it is difficult to make valid comparisons betwcen them that allow statements to be made about thc statistical significance of their difference. Use of proportional hazards modelling, however, permits a formal statistical comparison of different valves. This is a model that allows for the Significance of different prognostic factors to be considered. Each of these various factors can be looked at in turn so one can determine which onc accounts for the most variability in the response time. Thereafter, additional factors are included in the model to sce which, if any, improvcs it. This procedure is followed in a stcpwise fashion with all thc individual factors until a stage is reachcd when no further improvement to the model is possible. In this way, it is possible to use a large databank comprising many data on prognostic factors to determine which are the most important factors in relation to, for example, valve dysfunction, cvent-free survival or mortality. This analysis cannot bc achicvcd by the use of the life table/actuarial approach. Proportional hazards modelling has been used on Study data in an attempt to dctermine which preoperative factors could be rclatcd with time to valve dysfunction (valve dysfunction being cither valve failurc or infection, thromboembolism or haemorrhagc). As alrcady noted, the advantagcs of this typc of analysis over other methods arc that it allows the assessment of the effect of several preoperative factors acting simultaneously and provides an estimate of an individual’s projected time
to valve dysfunction given a composite profile of their preoperative characteristics. Figure 6 shows the estimated survival times for one set of prognostic factors. Each factor has a specific and significant effect on time to valve dysfunction. Many other similar plots can be produced by varying the pattern of these prognostic factors. For example, it may be more desirable to determine estimated survival times for patients with the same prognostic pattern, as shown in figure 6, except that they are aged over, not under 56 years. Figure 6 does, in fact, represent a slightly complicated model because rather than one survival function for a given set of prognostic factors there are five. T h e reason for this ‘stratified’ model is because the prognostic factor ‘valve type’ violated the assumption of proportionality. Proportionality means that any differences between the individual levels within a prognostic factor remain constant over time. Clearly, this is not the case here since the survival functions of disc, pig and tissue valves cross between 60 and 80 months. There are many possible subsets of the data which could be analysed using the proportional hazards model. Owing to the size of the Study such analyses are possible with little loss in power of the statistical tests employed. However, results of any such study using non-random data should always be treated tentatively. As with all statistical analysis based on the use on non-random data, care must always be taken in any interpretation involving behavioural prediction.
1 .o 0.9
z
t
0.8
U
5
6
f
0.7
0.6
E
8 81 f
0.5
c
O
&
0.4
a0
ng n
0.3
-A = -.- B C = --D = .----. E
=
Ball (558) Disc(2794) Pig (3026) lissue(747) Homograft(208)
\
\ \ \ \ \ \
\ \
\ \
\
A
\
\ Prognosticpattern D Age = Under 56 No previous embolism Not aortic stenotic No other operative diagnosis Aortic implant NYHC = 1
0.2 0.1
I
I
1
20
40
60
1
,
I
I
1
1
80 100 120 140 160 180 200
FOIIOW-up (months)
Figure 6. Proportional hazards analysis - estimated survival function. 7
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
P. J. D r u 7
ct
al. Valve data collection: problems and pitfalls
Comparison of cost-effectiveness of different types of heart valve
operation, cost of any ongoing therapy and cost of any subsequent valve-related event. The main benefit is a measure of survival after discharge.
The Study, as already noted, contains a wealth of data about almost 14,000 patients who have undergone heart valve replacement. Of particular importance is the availability of data concerning valve performance. Clinical performance of heart valves has been assessed fairly comprehensively but, to the best of the authors’ knowledge, no attempt has yet been made to perform a cost-effectiveness analysis of this field. Another aim of the Study, therefore, is to compare the relative costeffectiveness of using different types of heart valve. Though there are several types of heart valve, and within each type there are a variety of specifications and makes, this initial investigation focuses on a comparison between the two most commonly used types; mechanical and bioprosthetic valves. Later analyses will include different types and makes of valve.
Even though the valves can be considered to be direct substitutes in the vast majority of cases, their economic consequences are different (substantially). Mechanical valves, once fitted, can last for 20 years or more before re-replacement, whereas bioprosthetic valves last around 7 years. However, once fitted, bioprosthetic valves require no ongoing therapy whereas mechanical valves require some drug therapy, and so have a higher ongoing cost which counteracts the cost of more frequent replacement of the bioprosthetic valves. Figure 7 shows the two cost profiles (at current prices) over time. Both types of valve can be the cause of medical problems; events like thromboembolism and haemorrhage, infection and valve failure, and sometimes death. The Study has a record of such postoperative events and complications. These complications need to be treated and in all probability will require an inpatient stay. A significant cost will then be borne by the NHS. This cost must taken into account when comparing the costeffectiveness of the two types of valve. In order to estimate the additional cost imposed by each of these events, information is required about the likelihood of these events occurring. Using the database, the probability of any particular event occurring in any particular year after the initial operation can be estimated. Cost of the event in current prices, together with the probability of the event occurring, allows the expected cost of events in any given year to be included in thc cost profile. Thus, a cost profile (in current prices) can be built up to include operating costs, ongoing therapy and adverse events. However, costs relating to the two valve types are incurred at different times. Given that costs incurred in the future are worth less than costs incurred in the present, because of time preferences, some form of discounting needs to be introduced. Using the appropriate method of discounting enables expected lifetime costs to be built up for each valve type. These expected lifetime costs can be expressed in current prices, and thus are comparable directly.
The most appropriate method of carrying out an evaluation of this type is cost-effectiveness analysis. In a cost-effectiveness analysis, both costs and consequences of treatment in question are examined. It differs from simple cost or cost-minimization analyses in that it does not assume that the outcome differences of the two treatments are identical. O n the other hand, it does not require an explicit valuation to be made of the differing outcomes (if there are any differences) which a costutility analysis would require. A cost-effectiveness analysis allows the differences to be expressed in natural units (for example, life years saved) whereas a costutility analysis would require a valuation of those life years. Hence, no problem arises concerning assigning different valuations of life years to different groups of people. The two differing and most common forms of treatment that will be under consideration are valve replacements using mechanical and bioprosthetic valves respectively. It is proposed that homografts and other options will be included in a later study. Some patients may be treated with drug therapy as opposed to implantation, but this is a short-run option and these patients tend to be elderly or unsuitable for valve replacement. Generally, there is no ‘do nothing’ alternative; prospective patients cannot survive as long or with as good quality of life without a valve replacement. In addition, mechanical and bioprosthetic valves can be regarded as genuine options. There are only two significant exceptions. The first concerns patients who cannot tolerate anti-coagulant therapy. These patients will not be implanted with a mechanical valve, as ongoing anticoagulant therapy is required. O n the other hand, teenage and younger patients rarely receive bioprosthetic valves because of the higher risk of re-calcification. These two groups only account for a small number of patients though, for over 90% of operations there is a free choice between mechanical and bioprosthetic valves.
Cost per year
0
Given that a cost-effectiveness approach has been chosen, the next step is to identify the relevant cost and effectiveness data. These include the cost of the initial 8
5
Current prices
10
15
20
25
Year
Figure 7. Cost profiles of mechanical and bioprosthetic values.
P.J. Dmry rt
al. Valve data collrction: problems and pitfalls
Turning to effectiveness, it was established earlier that
Downloaded by [Universität Osnabrueck] at 21:59 14 March 2016
there was no acceptable ‘do nothing’alternative. Suitable patients are always implanted. Therefore, it becomes possible to estimate the number of life years gained by thinking in terms of survival after discharge. I n addition to simply thinking in terms of life years gained, there is also a crude facility within the database for measuring quality of life; the New York Heart Association classification, which is a method of measuring dysfunction. This can be used to build up a profile of quality of life as provided by dysfunction and it is hoped that this can be used in conjunction with the other data in the assessment of cost-effectiveness. Again, direct comparisons can be made beween the two types of valve. Thus, the Study database contains enough information for a crude assessment of cost-effectiveness. However, some assumptions have to be made, in particular the costs which will need to be obtained, the expected probability of adverse events and the appropriate rate of discount. A comprehensive sensitivity analysis is required to test the effects of varying these assumptions. This method has been designed for a basic study of mechanical and bioprosthetic valves. Methodology is very general and can be easily developed for use in other cardiac procedures. In terms of heart valves, it is envisaged that the work would be extended to include firstly homograft valves and then analysis of different makes of valve within general valve types. Methodology can be also refined further, for example, to include any alterations in the probability of additional events after a first event has taken place and to take account of age and sex in the selection of the most appropriate valve type. What has been presented in this section is a general framework for cost-effectiveness analysis applied to the comparison of different heart valves and utilizing existing data sources. This framework can be readily adapted for use in other fields and expanded to a cost-utility approach if the additional data required becomes readily available.
References 1 . BLACK,M. M., COCHRANE, T., DRURY,P. J. A N D LAWFORD, P. V. (1990) In vitro and in vivo performance of artificial heart valves Proceedings of the Conference on Medtcal and Biological Implant Technology, London, 27-29
March 1990. UK Liaison Committee for Services Allied to Medicine and Biology. 2. BLACK, M. M., COCHRANE, T., DRURY P. J. and LAWFORD P. V. (1987a) Artificial heart valves: past Performance and future prospects Cardiovascular Reviews and Reports. 8 , 40-45. 3. BLACK, M. M., COCHRANE, T., DRURY, P. J . and LAWFORD P. V. (1987b) Assessing the performance and safety of artificial heart valves. Proceedings of the 9th E M B S Conference, 3, 1183-1 184. 4. BLACK,M. M., DRURYP. J. and LAWFORD P. V. (1991) Measurement of the in vitro and in vivo performance of artificial heart valves. In A Concise Enqclopaedia of Biomedical and Biological Measurement Systems ( Pergamon Press, Oxford) p. 187-195. 5. BLACK,M. M., DRURYP. J. and SMITHG . H. (1983) Long-term assessment of heart valve substitutes. L f e Support System, 1 (Suppl. I ) , 301-304. 6. BLACK, M. M., DRURYP. J. and TINDALE, W. B. (1985) The clinical performance of bioprosthetic heart valves. In Biocompatibility of Tissue Analogs. Vol. 2 Edited by D. F. Williams (CRC Press, Boca Raton), pp 173-186. 7. BLACK,M. M., DRURY,P. J. and TINDALE, W. B. (1983) Twenty-five years of heart valve substitutes: a review Journal of the Royal Society of Medicine 76, 667-680. 8. DRURY, P. J., BLACK,M. M., LAWFORD, P.V. and KAY, R. (1987) The long-term clinical assessment of heart valve substitutes Engineering in Medicine, 16, 87-94. P. V. and BLACK,M. 9. DRURY, P. J., KAY,R., LAWFORD, M. (1986) Statistical reappraisal of the analysis of heart valve patient follow-up data-the estimation of valve failure rates LiJe Support Systems, 4, (Suppl. I ) , 121-123. 10. FESSATIDIS, I.TH, VASSILIADIS, K. E., MONRO, J. L., Ross, J. K., SHORE,D. F. and DRURY,P. J. (1989) Thirteen years’ evaluation of the Bjork-Shiley isolated mitral valve prosthesis. The Wessex experience. Journal of Cardiovascular Surgery, 30, 957-965. I I . WAIN,W. H. (1991) A unique porcine bioprosthesisthe Wessex experience. Revue Europeenne de Technologie Biomedicale, 13, 1 1 3 - 1 18. 12. WALESBY, R. (1983) A surgical assessment of the StarrEdwards mitral prosthesis. Current Medical Literature: Cardiovascular Medicine, 2, 65-67.
9
