NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Waugh N, Royle P, Craigie I, et al. Screening for Cystic Fibrosis-Related Diabetes: A Systematic Review. Southampton (UK): NIHR Journals Library; 2012 May. (Health Technology Assessment, No. 16.24.)

Cover of Screening for Cystic Fibrosis-Related Diabetes: A Systematic Review

Screening for Cystic Fibrosis-Related Diabetes: A Systematic Review.

Show details

5Health economics

Cost-effectiveness analysis is not possible at present due to lack of data, and so the purpose of this chapter is to consider modelling approaches, and to identify the data required.

Modelling approach

The model might follow 1000 children with CF, initially aged 10 years. This is based on guideline recommendations.

Arm 1 – Natural history/no screening

The base arm would be ‘natural history arm’ or NHA. There would be no screening, so there would be three groups of patients:

  1. those who do not develop diabetes
  2. those who do become diabetic but never have symptoms and are never diagnosed, so die earlier than they would have done had they been diagnosed and treated; note that some might die from unrelated causes and would not benefit from screening
  3. those who do develop symptoms, are diagnosed and treated, and live longer because of that.

The data required to populate Arm 1 are:

  • 1a What proportions of patients with CF develop diabetes at each age? Given that the natural history may be poorer in female patients, modelling should be done separately by gender.
  • 1b How many would be diagnosed because of symptoms? And hence how many would never be diagnosed without screening?
  • 1c How much longer do people with CFRD live once the diabetes is treated after diagnosis by symptoms? The fact that people with CFRD have shorter lifespans than those without may not be entirely due to the diabetes. It may be that more severe CF leads not only to diabetes, but also shortens life in other ways.
  • 1d What is the best treatment? Because the cause of the diabetes is loss of β-cells in the pancreas, and insulin sensitivity is normal, insulin is the standard treatment. But does that mean basal insulin, or mealtime boluses, or both, or CSII? The extra cost of CSII may be justified by improvements in QoL.

Figure 3 shows the outline of the model. The model shows the course the disease would take if a person were left untreated, unless diagnosed later owing to presenting symptoms. Without screening, people with CF could either become hyperglycaemic with symptoms and be diagnosed, could become hyperglycaemic without symptoms and remain undiagnosed, or could remain free of hyperglycaemia for the rest of their lives. There are different levels of hyperglycaemia:

FIGURE 3. No-screening model.


No-screening model.

  • PPH – for the purposes of this review, we define this as hyperglycaemia after meals, at 30, 60 or 90 minutes, but where BG level is normal by 2 hours: PG level of > 11.0 mmol/l at 30, 60 and 90 minutes, but is < 7.8 mmol/l by 2 hours.
  • IGT, where hyperglycaemia after meals has not returned to normal: FPG level of < 7.0 mmol/l and 2-hour PG level of ≥ 7.8 and < 11.1 mmol/l (WHO definition).
  • Diabetes mellitus.

In both of the first two levels, we assume that some would progress to the next level and others would not. Thus, the ones who develop IGT and are undiagnosed may either become diabetic or they may live with IGT until they die. Some of those patients that become diabetic will show symptoms and some will not. Those that do not show symptoms may live with diabetes, undiagnosed and untreated, till they die. Those that do show symptoms will be treated until they die.


Then we would have some screening arms. In each of these arms, we would need to model both longevity and QoL, to derive quality-adjusted life-years (QALYs).

Arm 2

Arm 2 would be the current screening default, the OGTT. This is usually only FPG and 2-hour PG, rather than the FOGTT. The baseline would be annual screening from the age of 10 years, but different thresholds could be examined. The key question might be when the benefits of treating detecting and treating those with diabetes are enough to justify the costs of screening, both in terms of monetary cost and inconvenience to those who do not have diabetes. Screening itself would have a disutility though this is transient.

Data required:

  • 2a How much longer do patients with diabetes live when it is detected by screening? Life-years gained.
  • 2b How good would their QoL in the added years be?
  • 2c Hence QALYs gained. Would some patients not live longer, but have better QoL after diabetes was treated? Some QALYs might be gained from QoL alone?
  • 2d What is cost of screening all patients once a year with OGTT? The cost will decline each year because those patients with diabetes will not need screened next year.
  • 2e What is the sensitivity of the test – would OGTT miss some patients? Although if screening is annual, they might only be missed for 1 year.
  • 2f What is the specificity of the test – would some patients be wrongly diagnosed with diabetes and treated inappropriately? (They would then probably get hypoglycaemia and be rapidly recognised as wrongly diagnosed, and have treatment stopped, so no long-term harm?)
  • 2g So far, we have not taken compliance into account. OGTTs are not popular, so not all patients would attend. Modelling has to take that into account, by adding a ‘screening-declined’ arm. It would start by assuming that those who decline screening have same outcomes as the NHA 1, but in practice, people who decline screening may have other health behaviours that make their outcomes poorer, so that might need a sensitivity analysis. So the screening arms all have two branches – those who accept and those who decline. The screening declined branch does not incur screening costs.

Compliance is important. A less sensitive but more acceptable test may result in more cases of CFRD being diagnosed. Oversimplifying:

  • OGTT 100% sensitive, but 50% acceptance identifies 50% of CFRD
  • HbA1c 80% sensitive, but 80% acceptance identifies 64% of CFRD.

We could then look at costs and benefits and consider whether or not screening with annual OGTTs is cost-effective.

Figure 4 shows the outline for modelling the screening arms.

FIGURE 4. Screening arms.


Screening arms.

Screening model

In this part of the model, it is assumed that patients with CF are offered screening, although they may not all accept. The acceptance rate may vary among the screening tests. Based on the evidence in earlier chapters, it is assumed that those patients who do not develop diabetes live longer than those who do. In addition, it is possible that patients who never develop PPH live longer than those who do, who live longer than those that develop IGT, who live longer than patients that develop diabetes. It is assumed that progression to diabetes is through PPH then IGT then diabetes (initially without FH and later with). It is assumed that screening and rescreening will take place every year from the 10th birthday onwards. With more data it may be possible to identify some patients whose risk is lower and who may be screened less often.

At the start of the model, screening is offered from 10 years of age. People who have a negative result are offered rescreening each year. Those who declined in the past are also offered screening. The negative results may contain both true- and false-negatives. From this screening point, people may have hyperglycaemia, of varying degrees, or they may not. If they never develop hyperglycaemia, when they die, it will be with CF alone.

If they do develop hyperglycaemia, they are treated. For simplicity, it is assumed that once a hyperglycaemic state is reached, regression to normoglycaemia does not occur, except for when transient hyperglycaemia is seen during acute infective exacerbations. As with the previous case, if the patients with PPH test negative for an IGT screen then they are rescreened the following year. If these patients remain stable and do not regress or progress, when they die, it will be with CF and PPH.

If the patients do develop IGT then they would be treated, if considered necessary. The key missing data here are whether or not earlier treatment (i.e. before diabetes has developed) is beneficial, as discussed in Chapters 2 and 3. The patients are monitored for progression rather than screened; they may remain stable with IGT, or they may progress and develop diabetes. Again, as with the previous case, if the patients test negative for diabetes, they are rechecked again the following year. If these patients remain stable and do not regress or progress, when they die, it will be with CF and IGT. If the patients that have tested negative are in fact false-negatives and if they were not screened again, or do not accept screening, they may at some stage develop symptoms and will be diagnosed and treated, or they may remain asymptomatic, undiagnosed, yet suffering harm. If the patients that have tested positive are true-positives and have developed diabetes, they are treated with insulin. These patients with CFRD do not regress and they do not progress to any other stage.

Other screening tests

We know that OGTTs are unpopular. Other screening options worthy of trialling are the 50-g non-fasting GCT (with dose adjusted for age or body weight) and CGM, so there should be an arm for each of those:

  • Arm 3 CGM
  • Arm 4 GCT
  • Arm 5 serial glucose profile undertaken at home: say 6–8 per day for 2 days.

Data requirements would be similar to the OGTT arm. All screen-positive patients would require a confirmatory second test, as asymptomatic diabetes should not be diagnosed on one abnormal glucose result.

It might also be worth having combination testing; for example, a first stage screen to reduce the number requiring OGTT.

What are we screening for?

In the present state of knowledge, it appears that screening for both CFRD and IGT would be worthwhile. We can hypothesise, based on Chapter 2, that it might be worth intervening as soon as patients start having episodes when glucose exceeds 8 mmol/l, but we have few data to support that at present.

Hence, the most important screening question at present is what we should be screening for.

Other problems with modelling

Survival with cystic fibrosis

Survival has been improving over recent decades, and we do not know how long those currently in the screening age band (assumed to be 10–30 years) will live for. We could use recent estimates from CF registries, and check these against the trends over time data from Dodge et al.,19 by extrapolating from the current survival lines for those diagnosed in recent decades.

Most papers on survival give mean age at death, but have a mixture of ages. We need data by decade of birth in order to estimate further improvements in survival.

The effect of cystic fibrosis-related diabetes

The figure of 11 years as being the loss of life-years owing to CFRD, based on the data from Milla et al.,57 could be used as the default, with other figures used in sensitivity analyses. However, we need three figures for loss of years:

  • those with diabetes detected by screening and treated at early stage
  • those with diabetes diagnosed and treated when they developed symptoms
  • those with undiagnosed diabetes – by definition data will not be available, but the 11-year figure could be used; however, this may be an overestimate, as the absence of symptoms may imply lower glucose levels.

Koch et al.14 found that patients with CFRD have a median survival age of 24 years compared with 34 years in non-diabetic control subjects with CF.

Several groups have reported a decline in clinical status occurs in patients with CF-related hyperglycaemia before the diagnosis of CFRD is made.15,54,122,146,206208 This may occur for several years before the diagnosis is made.

Overall annual incidence of cystic fibrosis-related diabetes

The overall annual incidence of CFRD was 3.5%209 but it will vary with age and one issue is when the incidence plateaus sufficiently for screening to stop, if indeed it does.

Quality-of-life studies in cystic fibrosis and cystic fibrosis-related diabetes

Health-related QoL has been described as ‘a multi-dimensional construct comprised of several domains as reported by the patient (e.g. physical, social and psychological functioning, respiratory symptoms, treatment burden and body image)’.210

To populate a model of CF and CFRD that allows us to evaluate the clinical effectiveness and cost-effectiveness of screening for CFRD, we ideally need:

  • Data on QoL over the lifetime of CF in patients who do not develop CFRD. We would expect a decline in QoL over the decades.
  • Similar data on those who develop CFRD but in whom it is not diagnosed, in whom we might expect a steeper decline in QoL.
  • Data on the QoL in those who are diagnosed and treated after developing symptoms. We might expect a diminution in QoL after onset but an improvement after treatment.
  • Similar data on those in whom CFRD was detected by screening. We might expect much less, or no, diminution in QoL before diagnosis, because the onset might be insidious, and by definition they would have no or few symptoms. However, they might feel better after starting treatment.
  • To assess the effect of treatment with insulin in patients with CFRD, and with lesser degrees of hyperglycaemia. For example, given that there is evidence that suggests that treatment with insulin may be of benefit at the IGT stage, we need to be able to quantify the effects on QoL.

One expectation might be that the development of CFRD would reduce the QoL, partly owing to symptoms or impaired performance, partly owing to the need for yet another treatment. A second might be that, in terms of QoL, the net reduction would be less in those detected early by screening, unless of course the disutility from insulin treatment was greater than the benefit from it, given that they are symptom free. The disutility will include that from injections, hypoglycaemic episodes and self-testing of BG level.

Ideally, we would have such data also for those with IGT, who would not include all the groups above, having no diagnosis via diabetes symptoms. However, they may still get benefit from treatment with insulin, and again there would be a trade-off between feeling better and the disutilities of insulin treatment.

We need both a sensitive measure of QoL that could pick up changes of value to people with CFRD, but also a generic measure of QoL from which we can derive a utility score for cost-effectiveness estimations.

We also need a measure that takes account of the fact that people with respiratory impairment may adjust their lifestyles accordingly.

Quittner211 reviewed the available instruments in 1998, dividing them into three main types:

  • Utility measures that provide a single value, with ‘1’ representing perfect health and ‘0’ representing death. They are used to generate QALYs and hence cost per QALY for assessing the cost-effectiveness of different treatments. Examples include the EQ-5D.
  • Health profiles, which generate scores for a number of domains of everyday living, such as energy, emotional state, physical functioning, etc. They can be applied to any disease state and hence may not be sensitive enough to detect disease-specific changes. Examples include the Short Form questionnaire-36 items (SF-36).
  • Disease-specific measures, designed to capture information on the symptoms and areas of functioning associated with specific diseases. They may therefore be more sensitive than health profiles, but cannot be converted to a generic utility measure. The main one discussed by Quittner is the Cystic Fibrosis Questionnaire (CFQ).

The CFQ consists of a suite of age-banded questionnaires, which include five generic domains (physical symptoms, role functioning, psychological/emotional functioning, energy and social functioning).

Studies with data on cystic fibrosis-related diabetes

Tierney et al.212 in Manchester compared QoL and experiences with hypoglycaemia in people with CFRD (treated with insulin) and T1DM. They noted that while there are studies in CF, there is a lack of studies in CFRD. Questionnaires were sent to 295 T1DM and 145 patients with CFRD. Instruments used included the Edinburgh Hypoglycaemia Scale (EHS) and the Diabetes Quality-of-Life (DQoL) measure. They noted that the DQoL had not been validated in CFRD.

Clinical data on HbA1c level, BMI and lung function (for patients with CFRD) were obtained from case notes.

The response rates were low: 52 (36%) patients with CFRD and 60 (20%) with T1DM. Of these, 20 patients with CFRD and 43 patients with T1DM completed diaries for hypoglycaemic episodes, giving return rates from the whole populations of 14% and 15%. The mean CFRD age was 30 years, and about half had CRFD for over 6 years.

Almost all patients had experienced at least one hypoglycaemic episode, but only 20% of the CFRD group had experienced hypoglycaemia with loss of consciousness, compared with 40% of the T1DM group. There was not much difference in hypoglycaemic episode symptoms, but the T1DM group reported slightly more neuroglycopenic symptoms.

Quality of life was better for the CFRD group than for the T1DM group: DQoL score 74 versus 66, respectively (a lower score is worse). This may relate to the hypoglycaemic episode scores on the EHS, which correlate with DQoL.

Reduced pulmonary function (FEV1) correlated negatively with DQoL.

Overall, the findings suggest that diabetes has less of a negative effect on QoL in CFRD than in T1DM, but the low response rates and inevitable bias should be taken into account. In addition, the authors suggest that, to people with CF, CFRD is just one more life-diminishing factor, and they retain some β-cell function, unlike those with T1DM. The CFRD group had fewer problems with hypoglycaemia than the T1DM group. They were less worried about the long-term complications of diabetes, perhaps because they had too many other problems to worry about, and perhaps because long-term diabetic complications have been less of a problem in CFRD and so receive less attention in clinics. This will change with increasing longevity.

There are around 20 studies of QoL in CF which do not mention CFRD. Brief details are given in Appendix 6. They fall into two main groups. First, there are those that use tools specific to CF, including the CFQ (five studies) and the CFQoL (one study). Second, there are those that use generic instruments, including:

  • Child Health Questionnaire (CHQ) (five studies)
  • Nottingham Health Profile (one study)
  • Quality of Well-Being Questionnaire (three studies)
  • EQ-5D (one study)
  • Sickness Impact Profile (one study)
  • Chronic Respiratory Disease Questionnaire (one study)
  • Questions of Life Satisfaction Questionnaire (two studies)
  • SF-36 (one study).
© 2012, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK98673


Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...