INTERCHEST clinical prediction rule for chest pain in primary care

Clinical background

Chest pain is one of the commonest reasons for adults to consult primary care, and for the general practitioner the central difficulty lies in separating patients with unstable coronary artery disease or another serious cause from those who can be managed in the community. Clinical assessment based on history and examination alone has proved insufficient to exclude coronary artery disease reliably, with a sensitivity of approximately 83 per cent for the general practitioner's unaided clinical judgement [4]. In practice this leads to a low referral threshold with many unnecessary emergency referrals, while individual patients with genuine coronary artery disease are nonetheless overlooked.

The INTERCHEST clinical prediction rule was developed to provide a structured, reproducible risk estimate that can be used during the consultation itself, without requiring an ECG or laboratory tests. Its purpose is not to diagnose coronary artery disease, but to identify patients whose probability is so low that emergency referral is not warranted, and to highlight those in whom suspicion is strong enough for immediate further investigation.

Calculating the INTERCHEST clinical prediction rule

The rule consists of six predictors, all equally weighted at +1 or −1, and the score ranges from −1 to +5:

$$\text{INTERCHEST} = \underbrace{I_{\text{age/sex}}}$${+1} + \underbrace{I{\text{serious condition}}}{+1} + \underbrace{I{\text{vascular disease}}}{+1} + \underbrace{I{\text{exertion}}}{+1} + \underbrace{I{\text{pressure}}}{+1} - \underbrace{I{\text{palpation}}}_{1}

where each $I$ is an indicator variable taking the value 1 if the criterion is met and 0 otherwise. The criteria are:

• Age ≥55 years (men) or ≥65 years (women): a sex-specific age threshold.
• The physician's initial impression that the patient may have a serious condition: the physician's own clinical intuition at first contact.
• A known history of coronary, cerebrovascular or peripheral arterial disease: previous ischaemic disease in any vascular bed.
• Pain brought on by exertion: typical exertional angina.
• The pain feels like "pressure": the patient's own description of its character.
• The pain is reproducible on palpation: the only negatively weighted criterion, which reduces the score by 1 when present, since reproducible chest wall pain argues against coronary artery disease.

The derivation cohort consisted of 3,099 patients from five primary care cohorts in five countries (Belgium, Germany, the Netherlands, Switzerland and Sweden), combined at the individual level through an individual patient data meta-analysis [1]. The prevalence of coronary artery disease in the cohort was 13.2 per cent. The outcome modelled was coronary artery disease confirmed at follow-up, and the predictors were identified through a combination of random forest trees, multiple imputation and logistic regression within each study, after which a common regression model was fitted to the pooled data allowing for systematic variation in missing data between studies.

Interpretation in practice

The threshold for decision is a score <2, which is the band in which coronary artery disease can be considered unlikely. In the derivation cohort this corresponded to a probability of 2.1 per cent (95% CI 1.1 to 3.9) [1]. At a score ≥2 the probability rose to 43.0 per cent (95% CI 35.8 to 50.4).

It is important to note that the score does not give a gradual rise in risk across the whole range but works primarily as a binary decision aid at the threshold of 2. Its clinical value lies in the high negative predictive value: a patient falling below the threshold can with considerable confidence be regarded as low risk, provided there are no other alarming features.

Validation and performance

In the derivation cohort a c-statistic of 0.84 was achieved [1]. The systematic review by Harskamp et al. summarised the diagnostic properties of INTERCHEST as a sensitivity of 82 to 88 per cent, a specificity of 74 to 82 per cent, a positive predictive value of 35 to 43 per cent and a negative predictive value of 96 to 98 per cent [2]. These figures come from the derivation cohort and a limited validation, and the review authors point out that INTERCHEST has not undergone as extensive external validation as the related Marburg Heart Score, which has been tested in several independent cohorts and shown a c-statistic of 0.84 to 0.90 [2].

The most comprehensive external evaluation of INTERCHEST to date was carried out by Manten et al. in a retrospective cohort study of 1,433 patients contacting out-of-hours primary care in the Netherlands in 2023 [3]. Here INTERCHEST was used as a standalone triage tool during telephone advice, with the triage assistant's assessment instead of the physician's as one of the predictors. The c-statistic for predicting serious events (a composite outcome including all-cause mortality and acute conditions requiring hospital care within 6 weeks) was 0.76 (95% CI 0.73 to 0.80), and for acute coronary syndrome specifically 0.77 (95% CI 0.73 to 0.81) [3]. At the threshold of ≥2, sensitivity was 81.7 per cent and specificity 61.6 per cent for serious events, and calibration was judged adequate [3]. Compared with the established Dutch triage protocol (NTS, c-statistic 0.66), INTERCHEST performed better and would have reduced the number of unnecessary referrals by 234 at unchanged safety [3]. For acute coronary syndrome specifically, sensitivity was 88.8 per cent and specificity 57.6 per cent, with fewer missed cases than NTS [3].

In summary, discrimination fell from 0.84 in the derivation to 0.76 to 0.77 in the external cohort. This may be explained in part by the study's use of the triage assistant's assessment instead of the physician's, by the retrospective design with a risk of verification bias, and by the broader outcome measure (serious events, not coronary artery disease alone).

Limitations

INTERCHEST was developed for, and validated in, primary care populations with chest pain. It must not be used in the emergency department or in patients with an already established acute coronary syndrome, in whom other instruments (HEART, TIMI, GRACE) and serial troponin sampling are indicated [2].

The systematic review flagged several quality shortcomings in the derivation study: more than 20 per cent of data were missing in parts of the cohort, and the risk of selection bias could not be fully excluded [2]. Nor has INTERCHEST been compared directly with unaided clinical judgement to the same extent as the Marburg Heart Score, for which such comparisons have shown that the score adds value beyond the physician's own judgement [4]. The review authors graded the evidence for INTERCHEST as level 4, which means that the rule is not yet considered ready for immediate clinical application and that further external validation is required [2].

One specific pitfall is that the predictor "the physician's initial impression" is subjective and can vary between observers. In Manten et al., the physician's judgement was replaced with the triage assistant's, which probably contributed to the lower discrimination [3]. This illustrates that the rule is sensitive to who makes the clinical assessment, and that it presupposes a physician with experience in primary care.

The rule is furthermore not designed to handle patients with an atypical presentation, such as isolated dyspnoea without chest pain, or patients with known coronary artery disease already on secondary preventive treatment who present with new symptoms. In these cases the score is not validated and should not be applied.

Derivation cohorts and ongoing primary care research

One of the five cohorts included in the derivation was Swedish, collected at Linköping University [1]. This gives the rule some direct relevance to Swedish primary care, even though external validation in Swedish practice is so far limited. The ongoing COSPRI trial in Region Östergötland, a cluster-randomised protocol for the investigation of suspected chronic coronary syndrome in primary care, illustrates Swedish interest in improving the diagnosis of chest pain at the primary care level, but does not use INTERCHEST as its intervention [5]. Swedish guidelines for the investigation of chest pain in primary care currently lack a formal recommendation on which prediction rule should be used, and the Marburg Heart Score has a somewhat stronger evidence base in the international literature [2].

References

1. Aerts M, Minalu G, Bösner S, et al. Pooled individual patient data from five countries were used to derive a clinical prediction rule for coronary artery disease in primary care. J Clin Epidemiol 2017. PMID: 27773828
2. Harskamp RE, Laeven SC, Himmelreich JC, et al. Chest pain in general practice: a systematic review of prediction rules. BMJ Open 2019. PMID: 30819715
3. Manten A, De Clercq L, Rietveld RP, et al. Evaluation of the Marburg Heart Score and INTERCHEST score compared to current telephone triage for chest pain in out-of-hours primary care. Neth Heart J 2023. PMID: 36580267
4. Haasenritter J, Donner-Banzhoff N, Bösner S. Chest pain for coronary heart disease in general practice: clinical judgement and a clinical decision rule. Br J Gen Pract 2015. PMID: 26500322
5. Nilsson S, Gabro F, Stertman E, et al. Chronic cOronary Syndrome in Swedish PRImary care (COSPRI): a study protocol for a 5-year cluster randomized controlled trial. Trials 2025. PMID: 40544312