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.)

Show details

# 4Systematic review of screening tests

## Terminology

The term ‘screening’ usually refers to the use of a simple but imperfect test, in asymptomatic people, in order to distinguish between those who probably have the condition and those who probably do not. It is usually used in the context of population screening but is also used in the context of screening people with a condition for a complication of it, such as retinopathy screening in diabetes. Screening tests are now being called ‘index tests’ in some research studies.

Those who have positive screening tests go on to a definitive diagnostic test, usually called the reference standard or sometimes ‘gold standard’, in research studies. The diagnostic test is assumed to more accurate and to give a definite diagnosis.

The reference standard test is usually more complex or more expensive; if not, it would be used as a perfect screening test.

Screening terminology includes the following terms, derived from the classic 2 × 2 table, as shown in Table 7.

#### TABLE 7

Classic 2 × 2 table for screening tests.

Sensitivity The per cent of patients with the disease who have positive screening tests. Those with the disease who are screening test-negative are false-negatives. Sensitivity = a/a + c.

Specificity The per cent of people who do not have the disease and who are screening test-negative. Specificity = d/b + d. So if specificity is 90%, 10% of people without the disease are screen-positives but false-positives.

Positive predictive value (PPV) = per cent of those with disease among those with positive screening tests a/a + b.

Negative predictive value (NPV) = per cent of those with a negative test who are true-negatives. It is about how good the screen test is at ruling out disease.

The reliability of a screening test can also be expressed as the per cent of results that are correct: a + d/a + b + c + d.

## Background

A survey in the USA by Allen et al.80 found a wide range of screening practices and tests for the detection of CFRD, with random PG the most common, followed by HbA1c, and urinary glucose. Most guidelines recommend an annual OGTT,172,173 but it appears that, owing to the cost, inconvenience and unpleasantness of the test, the guidelines are largely ignored in practice.80 Some of the variation in the tests used may relate to differences in the target diagnoses; tests may be perceived as being more or less able to detect different levels of glucose intolerance.

A survey in the UK obtained data from 37 of the 45 recognised centres (based on having ≥ 50 patients with CF).81 Most centres said that they screened patients annually. Most of the paediatric centres started screening at the of age 10 years, but a few started at the age of 12 years. The UK Cystic Fibrosis Trust recommends that screening should start at the age of 12 years.7

Six tests were used: the OGTT, random BG, serial glucose monitoring, HbA1c, FPG and glycosuria. It appears that the OGTT is the reduced version (ROGTT), with only fasting and 2-hour glucose levels measured, as recommended by the UK Cystic Fibrosis Trust, but the study does not say whether or not any units used the full OGTT (FOGTT). Serial glucose monitoring is taken to be a series of BG tests done with finger-prick, testing strips and meter; there is no mention of automated CGMSs being used. The commonest method used was the ROGTT, followed by various combinations of OGTT and other tests, such as FPG and HbA1c.

These methods may be the policies of the individual clinics, but what happens in routine care may differ owing to poor compliance. The survey did not provide data on numbers actually screened, and how.

### Issues

There is some evidence (see Chapter 3) that treatment may be beneficial not only in diabetes, but also in IGT. There is even a suggestion that treatment of isolated early PPH might be worthwhile, although this is based on very small numbers.

The suggestion of benefit from treating hyperglycaemia at non-diabetic levels would fit with the conclusion from Chapter 2, that adverse effects on the lung may start at PG levels as low as 8 mmol/l.

There are therefore uncertainties about what we should be screening for, with three groups:

• diabetes, including those without FH
• IGT

Given the transient nature of PPH, and the scanty evidence on benefit of treatment at that stage, we focus in this review on screening for diabetes, and for both diabetes and IGT.

Our default position is that diabetes and IGT are defined as per the WHO definition, but, as discussed in Chapter 2, this may be inappropriate if lung damage starts at lower levels of hyperglycaemia than retinopathy on which the WHO definition is based.

The screen-positives could potentially benefit in two ways – earlier treatment in those who would have been diagnosed later, after developing symptoms; treatment in those who would never have been diagnosed. We should also consider that some people who are detected and treated would never have developed symptoms and might have died from unrelated causes.

## Methods

### Criteria for considering studies for this review

#### Types of studies

Studies of screening tests can be:

• RCTs of one or more screening tests or strategies versus no/opportunistic screening.
• Case series, comparing a diagnostic test with an established reference standard. These can be either prospective or retrospective in nature.
• Case control, where test performance is compared between patients with known disease (i.e. diabetes) and those without the disease of interest; this type of design is known to be significantly more susceptible to bias than the case series design, especially when healthy control patients are included. The artificial selection of patients leads to an unrepresentative case mix.

Owing to the anticipated dearth of studies in the area, searches were for all study designs.

To be included for formal data extraction, studies had to report sufficient data for the construction of a 2 × 2 table.

#### Participants

Based on the findings of Chapter 1, it was decided that screening for CF-related hyperglycaemia would not start before the age of 10 years, and so studies of adults or children > 10 years were eligible for inclusion.

#### Reference standards

The test recommended by most consensus statements is the OGTT, often only in its reduced form. We assumed that the gold standard reference test is the FOGTT, but there are reservations about acceptability. However, we expected many studies to use the ROGTT as the reference standard, especially as the definition of diabetes is based on fasting and 2-hour results.

Reference standards for diabetes in CF therefore include:

• the 75-g (weight-adjusted) FOGTT result, with BG measured fasting and at 30, 60, 90 and 120 minutes
• the ROGTT, with only fasting and 2-hour measurements.

Ideally, a reference standard should indicate with absolute certainty the disease status of an individual. In reality, this is rarely achieved and less accurate reference standards must be accepted. For example, the ROGTT will miss PPH of the lag storage type, and even the FOGTT may miss hyperglycaemia if that occurs only in the evening. As will be reported later, there are also doubts about the reproducibility of the OGTT, so it is used more as a reference test than a gold standard.

#### Screening tests

Studies of any test to assess glucose intolerance in patients with CF were eligible for inclusion. These might include:

1. the 50-g glucose challenge test (GCT), with 60-minute glucose level
2. continuous glucose monitoring (CGM)
3. FPG
4. RBG levels
5. HbA1c
6. serial capillary blood glucose profiling
7. fructosamine
8. urine glucose tests
9. combinations of the above, for example a FPG test followed by an OGTT.

#### Search methods for identification of studies

As previously described in Chapter 3, a highly sensitive search strategy was run, in order to identify all aspects of patients with CF with diabetes and hyperglycaemia, including screening, diagnosis and treatment. Full details of the search strategy are shown in Appendix 1.

#### Selection of studies

Studies were selected for inclusion in the review in a two-stage process. In the first instance, the literature search results (titles and abstracts) were screened independently by two reviewers to identify all citations that appeared to meet our inclusion criteria as described above. Full manuscripts of all selected citations were obtained. One article in German174 and two in French175178 were translated into English. Where it was not possible to determine study eligibility from the title and/or abstract, the full manuscript was obtained. Any disagreements over study inclusion were resolved by consensus. It was never necessary to have arbitration by a third reviewer.

Studies were selected at two levels: first, those that yielded sufficient detail for 2 × 2 tables, and, second, other studies that might yield fewer but useful data.

The flow of studies is shown in Appendix 1.

#### Data extraction and management

For the first few studies, data were extracted independently by three or four reviewers, until we were happy that the predesigned data extraction form was satisfactory; some revisions were made. Information on study participants, study design, tests and reference test details, test performance (2 × 2 contingency tables) and potential sources of bias was extracted.

#### Assessment of methodological quality

The methodological quality of all included studies was appraised using a modified version of the QUADAS (quality assessment of diagnostic accuracy studies) tool.179 Ten items were initially included, but items 7a, 7b, 8 and 9 were deemed to be usually not applicable in a situation where results were numerical from a laboratory (and hence not susceptible to observer interpretation), and dichotomised. An 11th item on reporting of definitions of the different hyperglycaemic states was added.

Study quality was assessed by two reviewers. Each item was scored as ‘yes’, ‘no’, ‘unclear’ or ‘not applicable’. Appendix 3 shows the blank quality assurance form.

A summary of the reviews authors' judgements about the methodological quality item for each included study is shown in Table 8.

#### TABLE 8

Methodological quality summary.

Figure 2, presents a graphical summary of overall quality by showing the per cent of studies that did or did not fulfil each item.

#### FIGURE 2

The per cent of studies that fulfilled each quality item.

No summary scores estimating the overall quality of a study were calculated, as their interpretation is potentially misleading.180

The items of the QUADAS tool and their interpretation are as follows:

1.

Was the spectrum of patients representative of the patients who will receive the test in practice? The characteristics to be considered here included:

Age – the likelihood of diabetes increases with age, and so if the test was applied to a mainly older population it might appear more accurate. Hence, we looked for a sample of patients that was typical of the population in the centre, either paediatric or adult.

Selection bias, where we looked to see what proportion of the centre's patient population was included in the study. The greater the proportion, the less the bias. To estimate the proportion, we looked for the total clinic population.

Whether the patients on whom the screening test was being tested, had an over-representation of those with conditions likely to cause fluctuations in BG, such as exacerbations of lung disease. Studies in which all or a significant proportion were suffering from such exacerbations at the time of screening, were excluded.

Whether or not any particularly high-risk (or low-risk) groups were selected for screening.

2.

Is the reference standard likely to correctly classify the target condition?

For the reasons given above, we used the OGTT as the reference standard. Ideally, this would have been the FOGTT but the reduced version correctly classifies the target conditions (diabetes and IGT), as they are defined on the basis of it.

3.

Is the time period between reference standard and index test ≤ 1 month?

The time period between screening and reference testing needs to be short enough to ensure that the presence or absence of the condition does not change between tests. We assumed that a month (mean or median) was short enough, although this does leave some problems with skew. Ideally, we would exclude patients whose interval was much longer but studies did not give sufficient detail. In practice, it is probably more important that patients are in the same condition (e.g. free of infectious exacerbations) at both screening and reference testing.

4.

Did the whole sample or a random selection of the sample receive verification using a reference standard of diagnosis?

The issue here was whether the reference test differed according to the result of the screening test (e.g. if definite positives did not have the reference test but ‘borderline positives’ did).

5.

Did patients receive the same reference standard regardless of the index test result?

The issue here is whether all people having the screening test had the same reference test.

6.

Was the reference standard independent of the index test result (i.e. the index test did not form part of the reference standard)?

When OGTT is the reference standard, this does not apply to screening tests such as CGMSs or HbA1c. The FPG is part of the OGTT, but in practice, the diagnosis of CFRD is based more on the 2-hour level (because FH occurs later than PPH) and so this is not a problem.

7a.

Were the index test results interpreted without knowledge of the results of the reference standard?

Because of the objective nature of the screening and test results, neither this nor the next question were applicable.

7b.

Were the reference standard results interpreted without knowledge of the results of the index test?

8.

Were the same clinical data available when test results were interpreted as would be available when the test is used in practice?

Again, when the test results are objective and defined in advance, and not open to interpretation, this criterion is not applicable.

9.

Were uninterpretable/intermediate test results reported?

With objective testing, uninterpretable results should not be obtained. However, intermediate ones might arise if the investigators subdivided groups, for example splitting CFRD into those with and without FH, or into normal GT (NGT), IGT and diabetes. Problems would arise if results were described simply as normal or abnormal without defining meanings. Where intermediate results (usually IGT) were given, options included producing a 3 × 3 table, or two 2 × 2 tables, for example one defining abnormal as diabetes and normal as everything else, and the other defining abnormal as IGT + diabetes. Where appropriate, we used the second option, which seems correct given the possibility that treatment should start at the IGT stage.

10.

Were withdrawals from the study explained?

This usually refers to the possibility of bias if only some of the screened people go on to reference testing.

11.

Were definitions of the different hyperglycaemic states given?

This is important given changes in the classification of diabetes and other states, and differences in definitions such as the ADA and WHO definitions of IFG.

#### Data extraction

When data permit, 2 × 2 tables are produced for each study, with sensitivity, specificity, PPV and NPV, and CIs. Some studies report on IGT, and two 2 × 2 tables are produced: one with just diabetes as the target condition, the other with both diabetes and IGT.

### Analysis of 2 × 2 tables

Analysis was undertaken using the MedCalc diagnostic test evaluations program, version 11.6.1 (MedCalc Software, Mariakerke, Belgium).181

## Results

Nine studies174,183189,205 (one in German174) provided sufficient data for 2 × 2 tables with actual numbers, not just per cent, so that CIs could be produced. Full details are given in the data extraction forms in Appendix 4.

Studies are identified hereafter by the name of the first author and year of publication.

### Buck 2000

Buck et al.174,182 carried out their study in two hospitals in Ulm and Hannover, in 102 patients aged between 5 and 33 years, with a median age of 13 years. They compared the results of OGTTs (1.75 g glucose/kg body weight, up to a maximum of 75 g) with FPG and HbA1c levels. Results are shown in Tables 9 and 10.

#### TABLE 9

Buck 2000: screening for diabetes and IGT by HbA1c level.

#### TABLE 10

Buck 2000: screening for diabetes alone by HbA1c level.

Because the reference ranges for HbA1c level were slightly different in the two centres, results were pooled and reported as being normal or abnormal. The upper limits of normal were 5.0% and 5.7% in the two centres. It is not clear whether or not these limits were used to define screen positivity.

Of the 102 patients, 22% had IGT and 13% had diabetes. None of those with diabetes had experienced symptoms (perhaps because those with symptoms would have been diagnosed without screening), and none had an elevated FPG level. HbA1c level was not a sensitive test for diabetes.

### De Luca 1991

This Italian study by De Luca et al.183 included 39 children and adolescents, in the age range of 5 to 22 years, who had had normal random BG results over the previous year. Their BMIs ranged from 13 to 24 kg/m2. They had HbA1c tests and FOGTTs. Results were given for both diabetes (two patients) and IGT (seven patients); numbers were small. Insulin levels were also measured and noted to be normal when fasting, but delayed after the glucose load, even in some patients with normal OGTTs. The normal range for HbA1c level was 4–6%. The results are shown in Tables 11 and 12.

#### TABLE 11

De Luca 1991: screening for diabetes and IGT by HbA1c level.

#### TABLE 12

De Luca 1991: screening for diabetes alone by HbA1c level.

The authors commented, ‘In our experience, HbA1c did not constitute a sensitive and specific screening test for detection of patients with CF with glucose intolerance.’

The results were actually quite good, but with such small numbers, CIs were wide. There was no difference in HbA1c between patients with NGT and those with IGT.

### De Schepper 1991

De Schepper et al.184 from Brussels used HbA1c value of > 7.5% as the screening test in a group of 48 patients aged 2–28 years. All had a normal FPG (< 120mg/dl) and were clinically stable (which we take to mean absence of acute lung infection). They had FOGTT (but not full reporting of the intermediate results), which was considered abnormal if the 2-hour PG was > 140 mg/dl (7.8 mmol/). This was seen in 15 of the 48 patients. HbA1c level was over 7.5% in 22 patients (46%). It was normal in four patients with glucose intolerance, and 11 patients with normal OGTTs had raised HbA1c level. The results are shown in Table 13.

#### TABLE 13

De Schepper 1991: screening for diabetes and IGT with HbA1c level > 7.5%.

### Lee 2007

Lee et al.185 compared both the 50-g non-fasting GCT and HbA1c testing with the 75-g OGTT. Unfortunately, only just over half of those who had the GCT returned for the OGTT, and many did not do so within the intended 1-week period: the mean interval was 35 days but 61% returned within a week, and the mean is skewed by a 264-day outlier. The median was 7 days. The results are shown in Table 14.

#### TABLE 14

Lee 2007: screening for diabetes and IGT with 50-g GCT and HbA1c level testing.

The 50-g GCT had perfect sensitivity for diabetes and IGT. Most (six out of nine) patients had only IGT. The 11 patients who were OGTT normal but GCT abnormal had elevated 1-hour levels, which had returned to normal in the 2-hour OGTT. Abnormal was defined as PG level of > 7.8 mmol/l, but about half had results of > 11.0 mmol/l. Hence, the GCT appears to be useful for detecting PPH, which might cause alveolar fluid hyperglycaemia. Note also that the GCT was non-fasting, which could improve convenience. The authors conclude that the GCT is useful for reducing the number of OGTTs required, because none of the 35% of patients with normal GCTs had abnormal OGTTs.

It should be noted that information on postprandial glucose levels could also be obtained if the FOGTT was performed, but the GCT has the advantage of not requiring fasting.

### Magni 1996

Magni,186 from Italy, compared levels of HbA1c, FPG and PG 2 hours after breakfast. They also used fructosamine and glycosuria tests but found those unhelpful. Glycosuria was present in only two patients, one of whom had a normal OGTT. The recruits comprised 65 inpatients, but all were free of respiratory exacerbations and none was on steroids. The reason for admission is not given, but the implication is that they were admitted for assessment or research purposes. The results are shown in Tables 15 and 16.

#### TABLE 15

Magni 1996: screening for diabetes and IGT by HbA1c%, FPG and 2-hour PG post breakfast.

#### TABLE 16

Magni 1996: screening for diabetes by HbA1c and FPG levels.

The high sensitivities are not surprising in view of the low threshold because the thresholds were chosen to give complete capture, at a cost of poor specificity. Magni186 concluded that the ROGTT should be used as the screening test.

### Moreau 2008

Moreau et al.,187 from Strasbourg, compared the ROGTT with CGM in 49 patients. CGM involved 288 readings of tissue glucose per day. Four capillary BGs were required each day for calibration, so those could have been used as another screening option. However, no data were given in the paper.

For the OGTT, the standard WHO definitions were used to divide patients into NGT, IGT and diabetes groups. The CGM results were expressed in two main ways. The first was the presence of peaks of PG level of > 200 mg/ml (11.1 mmol/l). The second was quantitative: mean glucose value and AUC.

All patients with diabetes had peaks of > 200 mg/dl at least once after a meal, but so did 36% of patients in the NGT group and 52% in the IGT group. Results are shown in Table 17.

#### TABLE 17

Moreau 2008: screening for diabetes and IGT using CGMS peaks over 200 mg/100 ml.

The presence of the peaks in the NGT group may be due simply to some patients having PPH, and so rather than this causing a problem of false-positives it could be regarded as true-positives if it was decided that treatment was justified at that stage.

### Mueller-Brandes 2005

In one of the largest studies, Mueller-Brandes et al.205 used data from OGTTs in 1128 patients to assess the value of FPG alone. The FPG was at two levels, using the old and new ADA definitions for elevated glucose: ≥ 6.1 mmol/l and ≥ 5.6 mmol/l, respectively (Table 18). The authors' main question was whether in patients with FPG levels of < 5.6 mmol/l, OGTTs were unnecessary. In effect, the reference standard was the 2-hour PG, not the whole OGTT.

#### TABLE 18

Mueller-Brandes 2005 results on screening for diabetes and IGT using old and new ADA criteria for FPG.

Sensitivity and specificity were reported but no CIs were given, and it was necessary to read some figures from the graph to construct a 2 × 2 table, so what follows may not be very precise (Table 19).

#### TABLE 19

Mueller-Brandes 2005: screening for diabetes and IGT using old and new ADA criteria for FPG.

So, using the new 5.6 mmol/l threshold improves sensitivity but reduces specificity. However, even using the new ADA threshold for IFG, 18% of patients with diabetic OGTTs would have been missed. The authors conclude that FPG is unsatisfactory for screening for CFRD.

Mueller-Brandes et al.205 note that the OGTT is not a gold standard because of its poor reproducibility. They note the need for a confirmatory test but report that only 47% of those with a positive OGTT (34 out of 73 patients) had this confirmed by a repeat OGTT.

### Robert 1992

Robert et al.,188 from Paris, studied both FPG (> 6 mmol/l) and HbA1c (> 5.6%) levels as screening tests, with the FOGTT as the reference test, in a paediatric clinic. The mean age of the 49 patients was only 11 years, but the range was 2 to 21 years. The diagnosis of diabetes was based only on 2-hour levels of 11 mmol/l or above. Results are shown in Table 20.

#### TABLE 20

Robert 1992: screening for diabetes and IGT using HbA1c and FG.

Of 10 patients with glucose intolerance, seven were under the age of 10 years, with two aged 5 years.

### Yung 1999

Yung et al.189 investigated five screening tests and nine combinations of them, in 91 adult (> 16 years) patients attending the Royal Brompton Hospital CF clinic, London, UK, as shown in Table 21.

#### TABLE 21

Yung et al.'s results on screening for diabetes and IGT using a range of screening tests (reference test: diabetes vs IGT and NGT).

Based on the above findings, Yung et al.189 advocated a selective approach to screening, but because of their fairly small numbers, with only 12 patients with diabetes, they advocated larger studies.

## Other studies

A number of studies did not provide enough data for a 2 × 2 table but, nonetheless, provided some useful information.

Craigie et al. in the Royal Hospital for Sick Children in Glasgow have used BGP in children with CF, and have data (partly reported in conference abstract,75 partly unpublished) showing that home glucose profiling is more acceptable than the annual OGTT (so far, 100% acceptance of profiles vs 50% acceptance of OGTT) and had a number of advantages, including:

• It reflects ‘real-life’ situations, such as activities and meals.
• The technique is widely available and understood by all diabetes services.
• It does not require hospital attendance once the technique is taught.
• It is relatively inexpensive, for example compared with CGMSs.
• It is readily accepted by patients.
• It can be used to directly demonstrate the relationships between specific foods and BG.
• It provides multiple readings over a 24-hour period.

But, there are also some disadvantages:

• Waking is necessary to do overnight testing.
• There is not the same 24-hour profile as obtained with CGMSs.
• Capillary BG may be 10–15% higher than venous BG.
• The expense of the meter and testing strips.
• The need for repeated skin pricks.

One of the issues has been over the age at which to start screening. The Scottish Intercollegiate Guidelines Network (SIGN) guideline on diabetes (SIGN 116)190 recommends screening from the age of 10 years, as does the ISPAD guideline.171 However, a study from Naples by De Simone et al.,191 admittedly in only 22 patients, and available in abstract only, reported that 17% of patients below the age of 10 years had glucose intolerance. In a larger study, Ode et al.173 from Minnesota reported that 39 of 94 children aged 6–9 years had abnormal glucose tolerance (defined as either IGT or INDET). None had diabetes, but during 5 years of follow-up, CFRD developed in 42% of those with abnormal glucose tolerance at baseline and 3% of those with NGT.

The study by Dobson192,193 was carried out in two stages. First, FOGTTs were undertaken in 20 patients (originally 21, but one dropped out because of venepuncture problems). Five had IGT and were excluded from the next stage, which was a comparison with CGMSs. So the remaining 15 subjects all had NGT. They also had an equal number of control subjects without CF.

HbA1c, FPG and 2-hour PG levels were similar among the patients with CF with NGT and the comparison group, but those with CF had higher 30-, 60- and 90-minute PGs. Their mean CGMS level was also higher, by 14%. Five of the CF group had peak CGMS readings of > 11.1 mmol/l, compared with one of the non-CF group.

The value of this study comes from the demonstration that the CGMSs can detect PPH, whereas HbA1c level and the OGTT do not. If we link that with the (admittedly scanty) evidence from the pilot of treating at the PPH stage,138 the message may be that either CGMSs or intermediate levels after an OGTT could be the best test if we are to treat at PPH stage.

Franzese et al.194 also examined the use of CGMSs, this time in further investigation of PPH. Eighty-seven patients aged > 10 years had OGTTs, and 27 had at least one abnormal intermediate (30, 60 or 90 minutes, but details not given) level of > 7.6 mmol/l. Only this group, and five younger children who had experienced high glucose levels while on steroid treatment, had CGMSs. So, this was a study examining CGMSs only in subjects with previous PPH, rather than a screening study in a representative sample of patients with CF.

NGT, IGT and DM were defined by the 2-hour level of CGMS positivity by any value over a 72-hour period. The CGMS results classified more patients as having glucose intolerance than the ROGTT (Table 22).

#### TABLE 22

Comparison of ROGTT and 72-hour CGMS results.

However, it is likely that a FOGTT would have given similar results, as diabetes in the CGMSs is based on any one elevated glucose value over 72 hours. So this study does not show that CGMSs are superior to FOGTT.

Another small study of CGMS by Jefferies et al.,195 from Toronto, used CGMSs in a group of 19 adolescents who had all had at least one previous BG level of > 7 mmol/l (not clear whether blood or plasma). All of seven patients who were diabetic on OGTT 2-hour level were also diabetic by the CGMSs (> 11.1 mmol/l). The results for IGT were unclear: two of the six patients who had IGT by OGTT had NGT by CGMSs, and three of the seven patients who had IGT by CGMSs had NGT on OGTT.

O'Riordan et al.,196198 from Dublin, in a series of abstracts with increasing numbers, compare CGMSs and OGTT (FOGTT, because there is mention of five time points). HbA1c level was also measured. They assert that neither HbA1c nor OGTT are sensitive and advocate the use of CGMSs, but give insufficient details for 2 × 2 tables.

Middleton and Bishop199 (abstract only), from Sydney, reported that 17 of about 25 patients with abnormal OGTT, had normal HbA1c levels. They also repeated OGTTs 1–2 years later and noted regression to NGT in some (numbers not given).

Solomon et al.,200 from Toronto, also compared the results of the ROGTT with HbA1c and FPG in 10- to 18-year-olds, finding both insensitive. Of those with normal FPG levels, 17% had IGT, and 4% had CFRD. All of those with CFRD had pancreatic insufficiency, and there was an association with more severe classes of mutations. However, as the authors say, there is as yet no evidence that specific mutations predict CFRD. ΔF508 has been incriminated.

Thorsteinsson et al.78 (abstract only) provided insufficient data for assessing screening tests, but reported some useful natural history. The authors' key points were:

• At diagnosis of diabetes mellitus by annual OGTTs, FPG and HbA1c levels were raised in only 16% and 16%.
• Presence of IGT increased risk of later diabetes (odds ratio 5.6).
• But in 58% of IGTs, next OGTT was normal, so OGTT is far from a gold standard.

### The debate on the use of glycated haemoglobin

Iron deficiency is common in CF, and may be associated with higher HbA1c levels in people with T1DM.176 (Conversely, increased red cell turnover may be associated with reduced HbA1c level, and if present in CF could give a misleading indicator of glycaemia control.)

A small study from Texas by Hardin et al.201 (abstract only) divided nine patients with CF and previously detected IGT into those with good pulmonary function (FEV1 and FVC 82–92% predicted) and those with poor (FEV1 and FVC 32–48% predicted). Red blood cell turnover was faster in those with poor function, which led them to conclude that HbA1c level was not suitable as a screening test for CFRD.

Brennan et al.175 state that only about 10% of HbA1c comes from red blood cells surviving 80–120 days, suggesting that glycation is not linear, and that increased turnover would not necessarily affect the usefulness of HbA1c.

Allen202 notes the poor PPV of HbA1c (as reported by Lanng et al.33), advocating caution in its use as a screening tool in CF, and calling for a large trial.

Garagorri et al.,203 from Zaragoza in Spain, screened 28 patients with CF using HbA1c and FOGTTs. The authors say that the results of the OGTT were classified as per the WHO criteria. In total, 12 or 13 (the numbers are not entirely clear) had IGT or diabetes. HbA1c level was no different between the groups, suggesting that it was not sensitive enough to use as a screening test for IGT.

Holl et al.182 (letter only), from Hannover, also advised against the use of HbA1c level as a screening test, reporting a sensitivity of only 31% in 13 patients diagnosed with CFRD on the basis of 2-hour PG level of > 200 mg/dl. Insufficient data were given to derive sensitivity.

### Monitoring of glycaemic control in existing cystic fibrosis-related diabetes

Al-Aloul et al.204 (abstract only) examined the relative value of preprandial and postprandial PG in patients with known CFRD, being considered for insulin treatment. It is not clear how many patients their results were based on – the abstract says initially 11 but then mentions details for six. The main conclusion was that neither FPG nor HbA1c level was abnormal in most patients, but that the postprandial level usually was. No details are given of how the CFRD was diagnosed.

Brennan et al.175,176 set out to assess the value of HbA1c level in monitoring diabetic control in CFRD, and to compare its usefulness with monitoring in T1DM. They used CGMSs to determine mean PG. They compared the results in 20 people with CF, 10 of who had CFRD, with previous results from people with T1DM. They did not assess the value of HbA1c in screening for or diagnosis of CFRD.

They concluded that HbA1c level was a reliable guide to glycaemia in CFRD, the relationship between HbA1c level and mean BG level being similar to that in T1DM.

## Discussion

Is there an identifiable subgroup in which screening is not required? Is it possible to say that if people with CF do not have hyperglycaemia by, say, the age of 30 years, they will never get it? That implies that pancreatic damage ceases to progress. This is probably unlikely but there are clearly some people in whom CF is much less serious, although that might just mean they get complications such as diabetes much later in life?

The screening parameter relevant to this (hypothetical) subgroup would be NPV, which can be used to ‘rule out’ conditions.

Would combinations of tests give better results? Or provide a more cost-effective strategy, for example if a simple test could reduce the need for OGTTs in some patients, with only those with intermediate results going on to OGTTs.

One issue to be considered is acceptability. A strategy that is 90% sensitive and 90% specific but has 50% compliance, would detect 90 × 50 = 45% of true-positives. Specificity will always be 100% for the non-compliant (those who do not take the test can never be false-positives) so specificity will be 95%. If the most accurate test has lower acceptability to patients, other less sensitive tests with better compliance might in practice detect more cases.

## Conclusion

There is good evidence on tests that appear unsatisfactory, including HbA1c and FPG levels. There is less evidence on CGMSs, but it appears useful and may be especially so for detecting hyperglycaemia, which happens only at certain times of day. However, the diagnosis of diabetes is not based on elevations during CGM.

There is very little evidence on the 50-g GCT, but it may be the best test if the aim is to detect PPH. It can be given to non-fasting patients. However, as one of the (anonymous) referees pointed out, if the FOGTT is carried out, the intermediate values such as the 1-hour PG will also identify patients with PPH, and would have the advantage of linkage with the fasting and 2-hour values.

Meanwhile, guidelines continue to recommend screening for CFRD using the 75-g OGTT.172

Table 23 provides a summary of CFRD diagnostics studies.

#### TABLE 23

Summary of CFRD diagnostics studies.

A table of excluded studies, with reasons for exclusion, is shown in Appendix 5.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK98686