High-sensitivity troponin tests for the early rule out of NSTEMI

Summary estimates of sensitivity and specificity from testing a single sample, using a 99th percentile diagnostic threshold for the general population, were 90% (95% confidence interval [CI] 85 to 94) and 78% (95% CI 72 to 83) respectively, based on data from 22 studies. The summary estimates of sensitivity and specificity, using a single sample and a limit of detection threshold, were 99% (95% CI 97 to 99) and 36% (95% CI 28 to 45) respectively, based on data from 9 studies. The 8 studies that assessed the diagnostic performance of a limit of blank threshold in a single sample gave a similarly high summary estimate of sensitivity of 100% (95% CI 98 to 100), but a reduced specificity of 19% (95% CI 11 to 31). All estimates were similar when the analyses were restricted to studies that excluded people with ST-segment elevation myocardial infarction (STEMI).

Using multiple sample strategies appears to offer better specificity without substantial loss of sensitivity than using a single sample on presentation and a very low limit of detection or limit of blank threshold. The European Society of Cardiology's (ESC) 0/1‑hour rule-out pathway combines an initial sample and a very low limit of detection threshold (5 ng/litre) in patients reporting a minimum symptom duration of 3 hours. The strategy tests at presentation and 1 hour later for patients whose acute myocardial infarction is not ruled out by the initial test, that is, it uses an 'OR' combination. The threshold for repeat testing is an initial troponin concentration of less than 12 ng/litre with an absolute change in troponin concentration, from 0 to 1 hour, of less than 3 ng/litre. For NSTEMI, the sensitivity and specificity estimates for this strategy were 99% (95% CI 98 to 100) and 68% (95% CI 67 to 70), respectively. Estimates of diagnostic performance were similar for strategies using an 'AND' combination of initial high-sensitivity troponin level and absolute change in troponin level.

Summary estimates of sensitivity and specificity for a single sample using a diagnostic threshold of the 99th percentile for the general population were 75% (95% CI 65 to 82) and 94% (95% CI 94 to 96) respectively, based on data from 5 studies. These estimates were similar if the analysis was restricted to studies that excluded people with STEMI. The summary estimates of sensitivity and specificity, using a limit of detection threshold (2 ng/litre) in a single sample taken on presentation, were 100% (95% CI 99 to 100) and 21% (95% CI 16 to 26) respectively, based on data from 4 studies in NSTEMI.

For multiple sample strategies, the High-STEACS pathway was used as follows:

• in the whole population, symptoms for 2 hours or more and a troponin concentration of less than 5 ng/litre at 0 hours, or

• in women, 16 ng/litre or lower when measured 3 hours from presentation and an absolute change of less than 3 ng/litre at 0 to 3 hours, or

• in men, 34 ng/litre or lower when measured 3 hours from presentation and an absolute change of less than 3 ng/litre at 0 to 3 hours.

Using the High-STEACS pathway in this way appeared to offer increased specificity without substantial loss of sensitivity compared with a single sample on presentation and a risk stratification threshold of less than 5 ng/litre. The sensitivity and specificity estimates for this strategy were 99% (95% CI 97 to 100) and 76% (95% CI 73 to 78) respectively, for NSTEMI. The ESC 0/1‑hour rule-out pathway reported a lower specificity than the High-STEACS pathway. It used an initial sample and a limit of detection threshold of less than 2 ng/litre, or repeat testing combining an initial troponin concentration of less than 5 ng/litre and an absolute change in troponin concentration, from 0 to 1 hour, of less than 2 ng/litre. Summary sensitivity and specificity estimates were 99% (95% CI 98 to 100) and 57% (95% CI 56 to 59) respectively for NSTEMI.

The 2 studies evaluating the Access test each assessed a different multiple sample strategy. One followed the ESC 0/1‑hour rule-out pathway (initial sample and a limit of detection threshold of less than 4 ng/litre, or repeat testing combining an initial concentration of less than 5 ng/litre and an absolute change in troponin concentration, from 0 to 1 hour, of less than 4 ng/litre), giving sensitivity and specificity estimates of 99% (95% CI 94 to 100) and 70% (95% CI 66 to 74) respectively for NSTEMI. The second study assessed a similar strategy, but with repeat testing at 2 hours. The sensitivity estimates were similar for the 2 strategies, but the specificity of the 2-hour repeat testing strategy was higher than that of the 1‑hour strategy.

The study evaluating the VIDAS test assessed a multiple sample strategy, with samples taken on presentation and at 2 hours using a threshold of less than 2 ng/litre at presentation, or less than 6 ng/litre at presentation and at 2 hours. The reported sensitivity and specificity estimates were 98% (95% CI 92 to 100) and 64% (95% CI 59 to 68) respectively for NSTEMI.

The study evaluating the VITROS test assessed the ESC 0/1‑hour rule-out pathway. This combined an initial sample and a limit of detection threshold of less than 1 ng/litre, or repeat testing combining an initial troponin concentration of less than 2 ng/litre and an absolute change in troponin concentration, from 0 to 1 hour, of less than 1 ng/litre. The reported sensitivity and specificity estimates were 100% (95% CI 95 to 100) and 60% (95% CI 55 to 64) respectively for NSTEMI.

The study evaluating the TriageTrue test assessed the ESC 0/1‑hour rule-out pathway. This combined an initial sample and a limit of detection threshold of less than 4 ng/litre, or repeat testing combining an initial troponin concentration of less than 5 ng/litre and an absolute change in troponin concentration, from 0 to 1 hour, of less than 3 ng/litre. The reported sensitivity of this strategy was 100% (95% CI 97 to 100) and the specificity was 66% (95% CI 62 to 70) for NSTEMI.

Three studies evaluated the ADVIA Centaur test. Using a rule-out threshold of 2 ng/litre in a single sample taken on presentation, the sensitivity and specificity estimates were 100% (95% CI 99 to 100) and 23% (95% CI 21 to 25) respectively for NSTEMI. Two multiple sample strategies were evaluated. One followed the ESC 0/1‑hour rule-out pathway. This combined an initial sample and a limit of detection threshold of less than 3 ng/litre, or repeat testing combining an initial concentration of less than 6 ng/litre and an absolute change in troponin concentration, from 0 to 1 hour, of less than 3 ng/litre. The sensitivity and specificity estimates for this strategy were 99% (95% CI 95 to 100) and 56% (95% CI 52 to 60) respectively for NSTEMI. The second study assessed a similar strategy, but with higher thresholds and repeat testing at 2 hours. The sensitivity and specificity estimates for this strategy were 100% (95% CI 95 to 100) and 67% (95% CI 61 to 72) respectively for NSTEMI.

Using a rule-out threshold of 2 ng/litre, in a single sample taken on presentation, the sensitivity and specificity estimates for the Atellica test were 100% (95% CI 98 to 100) and 26% (95% CI 24 to 28) respectively for NSTEMI. Sensitivity and specificity estimates for the High-STEACS pathway (symptoms for at least 2 hours and an initial troponin concentration of less than 5 ng/litre, or a troponin concentration of 34 ng/litre or less in women, or 53 ng/litre or less in men at 3 hours and an absolute change in concentration from 0 to 3 hours, of less than 3 ng/litre) were 98% (95% CI 95 to 100) and 74% (95% CI 72 to 76) respectively for NSTEMI.

The study of the Dimension Vista test assessed a strategy using measurements done at baseline using a troponin concentration threshold of less than 5 ng/litre and an absolute change of less than 2 ng/litre within 1 hour. The sensitivity of the strategy was 100% (95% CI 97 to 100) and specificity was 66% (95% CI 62 to 69) for NSTEMI.

Estimates for the model input parameters were retrieved from the literature and from consulting experts. Accuracy estimates were derived from the systematic review component of the assessment. The proportion of people testing positive or negative was based on the estimated accuracy of the testing strategies considered (table 2) and the estimated prevalence of NSTEMI in the UK (12.2%).

Test-specific resource use consisted of the number of tests done and the duration of hospital stay in hours before discharge or acute myocardial infarction treatment. For test strategies that involved a subsequent test conditional on the outcome of the first test, the rule-out rate for the presentation sample was used to calculate the number of subsequent tests. The resource use included a delay from the time at which sampling could be done to the time at which results became available (2 hours) and delay between arrival at hospital and troponin assessment starting (1 hour).

Health state costs were taken from a retrospective cohort study done in the UK (Danese et al. 2016). Acute myocardial infarction treatment costs were based on NHS reference costs and hospital stay costs were based on data from the Personal Social Services Research Unit.

In the base case, test costs were assumed to be identical for all tests (£2.50) except for the point-of-care test (£25.00, based on information provided by Quidel). A scenario analysis was done using test-specific costs and assuming that costs relating to the analyser and staff time were identical for all strategies.

Age-dependent utility scores from the UK general population were calculated for people in the 'no ACS, no UA' health state. These age-dependent utility scores were combined with age-dependent disutility values for acute myocardial infarction, to calculate utility values for the 'post-AMI' health states (with or without reinfarction). Utility values for the 'UA' health state were calculated based on the 'post-MI' utility values and assuming a utility increment of 0.010.

The following assumptions were applied in the base-case analysis:

• Standard troponin testing (at presentation and after 10 to 12 hours) has perfect accuracy.

• Compared with acute myocardial infarctions occurring during the decision tree period, all acute myocardial infarctions (either first or reinfarction) occurring in the state-transition model are diagnosed correctly, so are treated.

• Unstable angina is always correctly diagnosed, so is treated.

• The reinfarction probability for the 'post-AMI with reinfarction' health state is equal to the reinfarction probability for the 'post-AMI' health state.

• The increased post-acute myocardial infarction reinfarction and mortality probabilities for untreated acute myocardial infarction were assumed to last 1 year. After this a relative risk of 1.0 was applied (for untreated compared with treated acute myocardial infarction).

• There is no additional benefit of starting treatment early, so treatment effect for high-sensitivity strategies is equal to treatment effect for standard troponin strategies.

• All 30‑day deaths (after presentation at the emergency department) are due to fatal acute myocardial infarction events and will receive the associated costs.

For the base case, it was assumed that people who tested negative on standard troponin tests and positive on high-sensitivity troponin tests would have a life expectancy and quality of life equal to people with true negative test results, but this assumption is debatable. A meta-analysis by Liplinski et al. (2015) showed that people with a negative standard troponin test and positive high-sensitivity troponin test have an increased risk of reinfarction and mortality compared with those who test negative on both standard troponin and high-sensitivity troponin tests. Although this risk was not as high as in people with both positive standard troponin and positive high-sensitivity troponin tests, it could still be considered prognostically important. A secondary analysis was done in which the risk of acute myocardial infarction and mortality was adjusted for people with false positive results.

In the base case, standard troponin testing was the most effective and the most expensive strategy. But other testing strategies with a sensitivity of 100% (subject to uncertainty) were almost equally as effective, resulting in the same QALY gain up to 4 decimal places. Compared with standard troponin testing, high-sensitivity troponin testing resulted in probabilistic incremental cost-effectiveness ratios (ICERs) ranging between £34,307 and £36,842,603 savings per QALY lost.

In the secondary analysis, standard troponin was the cheapest and the least effective testing strategy. Compared with standard troponin testing, high-sensitivity troponin testing resulted in probabilistic ICERs ranging between £4,043 and £6,148 per QALY gained.

For all scenario analyses of the secondary analysis, results were similar to those from the secondary analysis base case. Standard troponin remained the cheapest and the least effective testing strategy (deterministic analysis). In all scenario analyses of the secondary analysis, high-sensitivity troponin testing compared with standard troponin testing resulted in ICERs less than £10,000 per QALY gained.

One-way sensitivity analyses were done including all parameters that were changed in the probabilistic sensitivity analysis. In the secondary analysis, the parameters that had a notable effect on the estimated cost-effectiveness estimates were:

• 30‑day mortality for untreated acute myocardial infarction

• mortality 1 year after treated and untreated acute myocardial infarction

• discount rate used for outcomes

• relative mortality for people who had a true positive result compared with those who had a false positive result.
  
  At extreme values (based on 95% confidence intervals) of these inputs, standard troponin testing remains cheaper and less effective than the high-sensitivity troponin tests.