Statistics from Altmetric.com
Patients with cystic fibrosis (CF), their families, carers, insurers, health care planners, and CF carriers all have an interest in knowing the lifespan of people with the disease. Evidence-based medicine is now explicitly practised by many clinicians in their everyday clinical work. This practice should include prognosis,1 where the expected lifespan is the most important statistic.
However, clinicians with a responsibility for these patients are faced with a large literature on the survival of people with CF, which presents a contradictory picture. My purpose is to show how these contradictions may be resolved by reference to other published material. I have examined three “notable” observations to show what inferences may be reasonably drawn from them concerning the lifespan of people with CF.
In the absence of properly conducted randomised controlled clinical trials, observational methods have been used to try to determine the relative efficacy of different models of providing clinical care, even though such studies provide only weak evidence.2 In particular, the possible advantage of care at specialist centres compared to care by local paediatricians has been debated for many years. Because of the relatively small numbers of cases and local variations in the care delivered, international comparisons have been used to assess these two different strategies for care.
Basic epidemiological considerations
All data should relate to a well defined population, preferably the residents of a geographical region. Where a group is studied (such as people with CF) rules should exist which allow individuals to be allocated to that group (diagnosed) and members should be found by population screening. Case finding is less reliable than screening.
Two different ways of estimating survival
There are two different methods for calculating survival. Cohort survival would identify all people born with CF in a given time period; each subsequent year those surviving would be noted. Eventually all the cohort will have died and a complete picture of their survival will be available. This method takes a very long time. Estimates of median survival derived from such data are unlikely to be relevant to newly born cases because of improvements in treatment leading to a longer lifespan.
An alternative is the current survival method, which only requires observations over one year. All current cases alive in one year have to be identified and the deaths in that year noted. For each age, a mortality rate is calculated. It is then assumed that the calculated age specific mortality rates will apply to the current cohort over their future lifespan. Applying these mortality rates to the current cohort gives an estimate of their future survival. These data are the most up to date available, and represent a useful summary of the current age specific mortality rates. A fuller explanation of survival calculations is obtainable from many medical statistics textbooks.3 Caution is required when using such data to predict future survival, particularly for the CF population which has seen regular improvements in survival for the past 30 years.
EXAMPLE 1: AN EARLY ESTIMATE OF SURVIVAL
Over 30 years ago, life table data were given for children with CF attending one large hospital.4 This early quantitative data suggested that only a quarter survived to age 16 years for the period 1943–64. These data present such a striking contrast to normal population survival that any methodological weaknesses in the study cannot account for this difference.
There was little reported improvement in survival between those born in the first half of the period and those in the second. Although not stated explicitly, the results must have been obtained by current survival methods. Thus, this lack of improvement could be caused by survivor bias in the earlier group, with survivors being milder cases. The one year survival was given as over 80%, which is equivalent to that reported for the 1968–70 UK cohort.5 The survival to later ages in the 1968–70 cohort was much better than that of the 1943–64 group. Historical data for the UK show a continuing improvement in the mortality rate for all ages.5 This apparent lack of improvement in the survival of these infants compared to the improvement in all the other ages is consistent with an overestimate of the infant survival in the first group. Clinic based data only have cases that have survived long enough to be referred to the clinic. This effect can be seen in the relatively low mortality rates quoted for the first year of life in clinic based data compared to population based data. There is a smaller bias, in the opposite direction, caused by late diagnosis of adults (which still occurs) which would not be included in the data from a paediatric clinic. This early work correctly highlighted the survival consequences of CF. The weak methodology militates against using these results as a basis for measuring improvements in treatment. However, the survival of UK children with CF during 1943–64 was almost certainly worse than that reported.
EXAMPLE 2: AN INTERNATIONAL COMPARISON OF SURVIVAL TO EVALUATE PATIENT MANAGEMENT
A “comparison of the estimated (current) survival curves for CF in England and Wales, and Victoria, Australia for the years 1976–80” suggested Victoria was 18% better at 80% survival to age 20 years (fig 1).6 After eliminating respiratory infections and sudden infant deaths as possible explanations, attention is drawn to the fact that care for 90% of the children and adolescents in Victoria was provided by “a specialist centre” which was thought to be essential for optimum care.7 This care was contrasted with England and Wales where most children were looked after by their local paediatrician. The paper finished by suggesting the “need for further research in England and Wales into the reasons why the death rate... seems to be substantially higher than in Victoria”.
An independent estimate of UK current survival for the period 1980–85 gives only 55% survival to age 20.8
The current survival estimate for the period 1991–95 to age 20 for Victoria is 56% although a calculation which removed potential bias in the first few years of life gives 60%.9 The comparable UK figures are 64% for 1986–87,10 and 74% for 1994.5
These data could be taken to show that Victoria has seen survival to age 20 fall in 15 years from 80% to 56%. The UK, which still has 36% of its patients not attending a specialist centre,11 has seen survival to age 20 rise in 10 years from 55% to nearly 75%. These data span a period when many new therapies were becoming available for CF, and CF survival was universally regarded as improving.
These data, in total, present an implausible picture. Victoria is similar to the UK, with clinicians and other health care providers trained to the same exacting standards and a constant exchange of staff between the countries. The most likely explanation is bias in the early Victoria data. The number of deaths were small (26), so small fluctuations in the numbers dying could have a large influence on the data. The death data in the original paper, showing 20% mortality to age 20 years in Victoria, was most likely an underestimate of the longer term trend. Publication bias could be a contributory factor; the paper might not have been published had the results shown the opposite effect.
Contrary to the original claim,6 these data provide no evidence that the “specialist centre” of Victoria confers a better survival than the less centralised service provided in the UK.
EXAMPLE 3: NO RECENT IMPROVEMENTS IN SURVIVAL IN SOME COUNTRIES
However, data from Denmark show spectacular results with 80.4% survival to 50 years.14 The combination of these three sets of results has prompted the need for “a thoughtful review of current treatment programs and how they may be altered to improve the outlook for CF patients”.15
Both the latest Canadian and USA current survival curves show the same convex shape as the UK data (fig 2). This convexity is caused by successive cohorts having better survival than the previous. If the cohorts from North America are showing similar patterns to the UK data, then improving survival must be inevitable for the next few years at least. This is confirmed by the best estimates that can be made of the annual CF births and deaths. For both countries, births per year are consistently about double the deaths. This can only happen when survival is increasing, there being no suggestion of any change in the incidence of the disease. The apparent ending of improvement in survival is a combination of chance fluctuations in the numbers dying and probably some short term corrections for overestimated survival in previous years because of the weak definitions of the populations under study.
The Danish data are just not plausible. The survival curve estimates no deaths for people with CF between the ages of 35 and 50 years of age! These estimates are very imprecise, being based on 270 patients and on what appear to be four deaths. Before there is a need for “a thoughtful review of current treatment programs...” based on these data, it would be better if the data were thoughtfully reviewed.
In large populations, there is no evidence that one country with a well developed knowledge and treatment program for CF has a markedly better survival than any other similar country, and thus no evidence that one country’s method of delivering health care to people with CF is “the best”.
The perplexing issue of the expected survival of a newborn baby with CF remains unresolved as there are insufficient data to make a reliable estimate. Recent cohort data5 record dramatic improvements in survival over the last 30 years. Attempts at extrapolating these improvements to predict survival over the next 50 years or more are fraught with uncertainty. The latest cohort data from the UK show that those born in the 1990s have a survival which is close to that of the overall population. For CF cases born in 1995 (about 300) there have been two deaths reported in the first year of life. At best, the survival of CF cases might track that of the base population. At worst, there might be some unexpected complication, which, in the future, will halt these improvements resulting in a median survival below 30 years.
A guess as to the expected lifespan by a supposedly reputable source, based on an analysis with questionable assumptions, will almost certainly be wrong and risks disadvantaging one group or another. Insurers, health care planners, patients, their families, and CF carriers all have an interest in knowing the lifespan. It is better to acknowledge that this is unknown, and plan with that uncertainty, rather than take decisions on incorrect data.