Use of multiple inflammatory marker tests in primary care: using Clinical Practice Research Datalink to evaluate accuracy

Background Research comparing C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), and plasma viscosity (PV) in primary care is lacking. Clinicians often test multiple inflammatory markers, leading to concerns about overuse. Aim To compare the diagnostic accuracies of CRP, ESR, and PV, and to evaluate whether measuring two inflammatory markers increases accuracy. Design and setting Prospective cohort study in UK primary care using the Clinical Practice Research Datalink. Method The authors compared diagnostic test performance of inflammatory markers, singly and paired, for relevant disease, defined as any infections, autoimmune conditions, or cancers. For each of the three tests (CRP, ESR, and PV), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under receiver operator curve (AUC) were calculated. Results Participants comprised 136 961 patients with inflammatory marker testing in 2014; 83 761 (61.2%) had a single inflammatory marker at the index date, and 53 200 (38.8%) had multiple inflammatory markers. For ‘any relevant disease’, small differences were seen between the three tests; AUC ranged from 0.659 to 0.682. CRP had the highest overall AUC, largely because of marginally superior performance in infection (AUC CRP 0.617, versus ESR 0.589, P<0.001). Adding a second test gave limited improvement in the AUC for relevant disease (CRP 0.682, versus CRP plus ESR 0.688, P<0.001); this is of debatable clinical significance. The NPV for any single inflammatory marker was 94% compared with 94.1% for multiple negative tests. Conclusion Testing multiple inflammatory markers simultaneously does not increase ability to rule out disease and should generally be avoided. CRP has marginally superior diagnostic accuracy for infections, and is equivalent for autoimmune conditions and cancers, so should generally be the first-line test.


INTRODUCTION Aim
To compare the diagnostic accuracies of CRP, ESR, and PV, and to evaluate whether measuring two inflammatory markers increases accuracy.

Design and setting
Prospective cohort study in UK primary care using the Clinical Practice Research Datalink.

Method
The authors compared diagnostic test performance of inflammatory markers, singly and paired, for relevant disease, defined as any infections, autoimmune conditions, or cancers. For each of the three tests (CRP, ESR, and PV), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under receiver operator curve (AUC) were calculated.

Results
Participants comprised 136 961 patients with inflammatory marker testing in 2014; 83 761 (61.2%) had a single inflammatory marker at the index date, and 53 200 (38.8%) had multiple inflammatory markers. For 'any relevant disease', small differences were seen between the three tests; AUC ranged from 0.659 to 0.682. CRP had the highest overall AUC, largely because of marginally superior performance in infection (AUC CRP 0.617, versus ESR 0.589, P<0.001). Adding a second test gave limited improvement in the AUC for relevant disease (CRP 0.682, versus CRP plus ESR 0.688, P<0.001); this is of debatable clinical significance. The NPV for any single inflammatory marker was 94% compared with 94.1% for multiple negative tests. request) as well as linked data from the English Cancer Registry for cancer codes.

Index tests
The index tests were CRP, ESR, and PV; test results were dichotomised into raised or normal using the mean upper limit of normal from laboratories within this study (>7 mg/L for CRP; >1.72 mm/hour for PV; upper limits of normal, stratified by age and sex, for ESR are available from the authors on request). A binary variable 'any raised inflammatory marker' was generated if any of CRP, PV, or ESR were raised.

Accuracy of CRP, PV, and ESR as single tests
For each of the three tests (CRP, ESR, and PV), dichotomised test results were cross-classified with the reference standard 'any relevant disease', allowing sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) to be calculated. Logistic regression was used to calculate diagnostic odds ratios, with and without adjustment for age and sex.
To address potential concerns that differences in patient mix could lead to biased estimates, for example, CRP used preferentially in patients with suspected infection, sensitivity analyses were conducted on the subgroup with two tests performed simultaneously to allow head-to-head comparison of diagnostic test accuracy.
Test results were also treated as continuous variables on a log scale, owing to their skewed distribution, to assess their predictive value in a logistic regression model, including age and sex as additional explanatory variables, calculating the area under curve (AUC). The AUCs for CRP versus ESR plus CRP versus PV were compared using the DeLong method, 9 generating confidence intervals (CIs) and P-values. Sub-analyses compared AUCs for disease subtypes, including infections, autoimmune conditions, and cancers.

Accuracy of test results in combination
The authors examined the accuracy of two combinations of inflammatory markers: CPR plus ESR and CRP plus PV. Only 111 patients had ESR plus PV and 306 had all three tests (Figure 1), therefore the researchers did not examine these test combinations. Where two inflammatory marker tests were performed simultaneously, measures of diagnostic accuracy were calculated (sensitivity/ specificity/PPV/NPV) for two alternative definitions of an overall positive result: • both inflammatory markers raised (denoted, for example, as CRP + ESR or CRP + PV) -defined as a combined test where both inflammatory markers tested were positive; and • either inflammatory marker raised (denoted, for example, as CRP|ESR or CRP|PV) -defined as a combined test where either of the inflammatory markers tested were positive.
The AUC for test combinations were

How this fits in
There is a lack of research comparing the accuracy of inflammatory markers. Testing multiple inflammatory markers is common, leading to concerns about overuse. In this large observational study using UK primary care electronic health records the authors found very little difference between the accuracy of C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), and plasma viscosity (PV). CRP had slightly superior diagnostic accuracy for infections, and was equivalent for autoimmune conditions and cancers; the authors therefore suggest this should be the first-line test in most circumstances. Testing multiple inflammatory markers does not increase the ability to rule out disease and should generally be avoided. calculated using a logistic regression model with log-transformed test values, including age and sex as covariates. An interaction term was used in the model due to the associations between inflammatory marker test results. All analyses were carried out using Stata (version 15).  Figure 1

Test results and overall disease incidence
In the single test group 23.8% (n = 19 932) had a raised inflammatory marker (Table 1). In comparison, 34.0% (n = 18 078) of the multiple test group had one or more raised inflammatory marker; 12.8% (n = 6803) had concordant raised values and 21.2% (n = 11 275) had discordant results (one raised, one normal). Overall disease incidence was 8.2% in the single tested group compared with 9.0% in the multiple tested    Table 3). The authors found no significant difference in the AUC of CRP and ESR for diagnosis of autoimmune conditions, and no significant difference for the main subtypes of autoimmune disease: polymyalgia rheumatica, rheumatoid arthritis, seronegative arthritis, or inflammatory bowel disease.
On comparing CRP and PV it was found that CRP had a higher AUC for infection (AUC 0.638, 95% CI = 0.608 to 0.670 versus   Table 2).
If an overall positive result was defined as both inflammatory markers raised, for example, CRP plus ESR, then PPVs were higher and specificity was increased, but at the price of lower sensitivity, compared with using any single test. If the combined test was defined as either inflammatory marker raised, for example, CRP|ESR, then sensitivity increased but specificity fell compared with any single test. This led to fewer false-negatives or reduced risk of missed diagnoses but a markedly increased frequency of false-positives, for example, CRP alone generated falsepositives in 19.3% of those tested, compared with 32.5% false-positives for CRP|PV ( Table  2, column 4). The maximum sensitivity was 60.6% for the test combination CRP|PV ( Table 2, column 7).

Accuracy of test results in combination: area under curve (AUC)
The authors compared the accuracy of CRP and ESR in combination, compared with the better of the two individual tests (Table 3). Adding a second test gave limited improvement in the AUC for relevant disease (CRP 0.682, 95% CI = 0.672 to 0.690 versus CRP plus ESR 0.688, 95% CI = 0.678 to 0.697, P<0.001). There was no improvement in AUC for CRP and ESR in combination for infection. The combined test CRP plus ESR gave an increase of 0.014 in the AUC for autoimmune disease (P<0.001) and 0.003 increase in AUC for cancers (P = 0.006) compared with single CRP test. While this was statistically significant, it is unlikely to be of a magnitude to be clinically significant. The combined test did not increase the AUC for polymyalgia rheumatica, seronegative arthritis, or inflammatory bowel disease, and led to a small increase of 0.009 in the AUC for rheumatoid arthritis (P = 0.007).
Similarly, the combination of CRP and PV together gave no improvement in AUC, compared with the better of the two individual tests, for infection or cancer (Table 4). The combined test CRP and PV gave an increase of 0.022 in the AUC for autoimmune disease, which was statistically significant, but seems unlikely to be clinically significant.

DISCUSSION Summary
In this large study of UK inflammatory marker testing, the authors found the practice of requesting multiple inflammatory markers to be remarkably common, perhaps reflecting increases in overall primary care testing rates. 2 Multiple testing was associated with more abnormal and more discordant results. The authors found no evidence that this approach helps to rule out serious pathology, as the NPV of a single inflammatory marker (94.0%) was the same as the NPV of combined inflammatory markers (94.1%) ( Table 1). Furthermore, discordant results may be challenging to interpret, critically, whether the clinician should regard one abnormal test as sufficient or should require both to be abnormal before further investigation or treatment. No combination of inflammatory marker tests can be used to rule in or rule out disease confidently. The maximum sensitivity of 60.6% (for the combined test CRP|PV) is low, yet comes at a price of increased false-positives compared with using single tests.
In diagnosis of infections, CRP marginally outperforms both ESR and PV. The three tests are equivalent for diagnosis of autoimmune diseases and cancers. Overall, inflammatory markers have a low AUC for most disease outcomes, with the exception of polymyalgia rheumatica. Testing multiple inflammatory markers, perhaps unsurprisingly, produces a higher PPV if both tests are raised (22.6%) compared with a single raised inflammatory marker (PPV 15.0%). This benefit is offset however by the low sensitivity once a double positive result is required (32.1% for CRP plus ESR and 32.7% for CRP plus PV), meaning that more pathology would be missed with this testing strategy, making it less helpful for ruling out disease. Testing two inflammatory markers does not appear to improve the overall discriminatory ability measured by the AUC. The small differences in AUC between single and double inflammatory marker tests for autoimmune conditions and cancer are probably of little clinical value, even if statistically significant.
Testing multiple inflammatory markers simultaneously does not improve the ability to rule out disease. It leads to increased rates of discordant results and increases costs without tangible benefits. CRP should generally be the first-line test, with the possible exception of myeloma. It must be remembered that all inflammatory markers have relatively poor performance characteristics, so perhaps is it no surprise that two tests are no better than one.

Strengths and limitations
The major strengths of this study are its size and the setting in primary care, where the initial suspicion of disease usually arises. Given the large sample size, the researchers have been able to directly compare diagnostic accuracy in patients with two inflammatory markers performed simultaneously. This reduces the potential for selection bias, where tests might perform better for certain disease outcomes due to GPs pre-selecting those at higher risk to have a specific test, for example, preferentially using CRP when an infection is suspected. There the possibility remains that patients with multiple inflammatory markers may differ from those with a single test; this is reflected by the fact that overall rates of disease were 9.2% in the double-tested compared with 8.2% in the single-tested group. This may influence the generalisability of the present results; however, the finding that measures of diagnostic accuracy are very similar in sensitivity analyses limited to the double-tested groups suggests that this is a relatively minor effect. One complication of a large sample size is that statistically significant differences can be found that are of little clinical significance; the authors have tried to highlight where this occurred.
Using routine data for diagnostic accuracy studies rather than prospectively performing multiple tests and evaluating a single disease outcome more closely reflects the diagnostic dilemmas facing GPs; however, this innovative approach does bring inherent challenges. The authors chose to use 1-year incidence of cancer and autoimmune disease, and 1-month incidence of infection, as a proxy for prevalence of disease at the time of testing. This is a pragmatic choice, based on evidence of the time lag between symptomatic presentation and diagnosis of cancer, 10 but, as a result, some of the diagnoses may be unrelated to the initial inflammatory marker test result.
All studies using electronic health records are reliant on the quality of data recording; however, blood tests are transferred electronically into the notes and diagnoses tend to be recorded with greater accuracy than symptoms. 11 The authors also used cancer registry data to improve cancer outcome ascertainment. Diagnosis of infection is likely to be less well coded, and microbiological confirmation of diagnosis is rarely obtained, leading to potential biases.
In clinical practice several factors determine who is tested: the patient, the symptoms, and the GP. Though the researchers have data on the demographics of the patients tested, they do not know what symptoms triggered testing and therefore cannot determine which tests were done for specific diagnostic purposes, and which were done as a general rule-out test for any relevant underlying disease. Demographic characteristics of GPs may also influence the choice of inflammatory marker test used; however, the researchers did not have GP identifiers so were not able to explore potential clustering by GP.
The benefit of the present approach is that it reflects real-life clinical practice; though GPs may not have a specific diagnosis in mind when they request inflammatory markers, they need to consider a wide range of possible diagnoses if the test is positive.

Comparison with existing literature
In a previous systematic review, limited evidence comparing CRP and ESR was found for a small number of specialist disease outcomes in secondary care settings. 4 The limited evidence available prevented the authors from making recommendations about the preferred choice of test. The PPVs in this study are lower than those reported in that review; this is likely to reflect the low disease prevalence in the primary care setting.
UK guidelines for diagnosis of polymyalgia recommend measuring ESR and CRP; 12 however, the authors were not able to demonstrate an improvement in diagnostic accuracy from combining these two tests. Previous studies have shown that inflammatory markers can sometimes be normal in both polymyalgia and giant cell arteritis; 13,14 in another study, most of those with normal ESR had raised CRP. 15 In cases of diagnostic uncertainty, repeat testing is often warranted, a different inflammatory marker may be added at this stage, or the same test repeated, expecting a change over time. The authors were unable to examine this.
Previous studies have shown that ESR and PV are superior to CRP for the diagnosis of myeloma. 16 Due to the small number of myeloma cases in the present sample the researchers were unable to corroborate this finding.
Although the authors have been able to show moderate predictive value of inflammatory markers for inflammatory bowel disease (IBD), the AUC for CRP of 0.698 (in a model that includes age and sex) is much lower than for calprotectin with a published AUC of 0.95, 17 therefore calprotectin is to be preferred if IBD is under consideration. Similarly, though inflammatory markers have a modest AUC for rheumatoid arthritis, low sensitivities found in the present study are in keeping with previous studies, which have found that 35% to 45% of patients with rheumatoid arthritis have normal inflammatory marker levels at diagnosis; 18 National Institute for Health and Care Excellence guidelines therefore recommend referral of patients with clinical evidence of rheumatoid arthritis, even with normal inflammatory marker test results. 19 It is therefore hard to see any benefits from inflammatory marker testing where rheumatoid arthritis is suspected diagnostically, though it may have a useful role in disease monitoring.

Implications for practice
Testing multiple inflammatory markers does not improve the ability to rule out disease, but does increase the risk that at least one of the tests will give a false-positive, compared with a strategy of using a single test. The authors therefore suggest that this should generally be avoided, in keeping with primary care guidance in New Zealand. 20 The overall diagnostic utility of all three inflammatory markers is similar and low, however CRP marginally outperforms ESR and PV for infections. CRP also tends to be cheaper than either ESR or PV (1.19 GBP for CRP, 3.18 GBP for ESR, 3.18 GBP for PV: source Bristol North Somerset and South Gloucester CCG laboratory costings). The authors therefore suggest that CRP should be the first-line test in most circumstances. Exceptions might include the use of ESR or PV rather than CRP for suspected myeloma (given that the authors have no evidence to support or refute previous findings), though if there is strong clinical suspicion then direct testing using electrophoresis and Bence Jones protein is preferable.
There is no combination of inflammatory markers that can be used as a reliable rule-in or rule-out test strategy. Results and decisions to test must be made in the context of other clinical findings. Faced with low probability of disease, for example, 'lowrisk-but-not-no-risk' cancer symptoms, inflammatory markers may still offer some clinical utility. They should however be interpreted in a Bayesian manner, with a positive test result increasing disease likelihood, and a negative test reducing disease likelihood, with neither being definitive. However, a negative test in the clinical context of a low-likelihood situation may be sufficient to provide reassurance.

Funding
This report is independent research arising from Jessica Watson's doctoral research fellowship (reference number: DRF-2016-09-034) supported by the National Institute for Health Research (NIHR). This research is linked to the CanTest Collaborative, which is funded by Cancer Research UK (reference number: C8640/A23385), of