C-reactive protein and clinical outcome in COVID-19 patients: the importance of harmonized measurements

Introduction

Since the early phase of the SARS-CoV-2 pandemic, many studies have pointed out how elevated C-reactive protein (CRP) concentrations in serum were highly prevalent in individuals affected by coronavirus disease 2019 (COVID-19), also correlating the entity of elevations with disease severity and patient outcome. Pooled prevalence rates of elevated CRP in COVID-19 patients varied in different publications between 60 and 80 %, with some individual studies showing increases in more than 90 % of patients [1], [2], [3], [4]. In a retrospective study performed by our group on a population of 2054 hospitalized COVID-19 patients, the prevalence of elevated CRP (>10 mg/L) was of 91 % [5]. From a pathophysiological point of view, CRP elevations in COVID-19 patients are a direct reflection of the immune hyperactivation caused by SARS-CoV-2 infection, which in some cases may evolve in a full-blown cytokine storm, with severe to life-threatening systemic consequences for the patient [6].

In Table 1 we have summarized the findings of meta-analyses published throughout the pandemic, investigating the role of CRP in the prognostic classification of patients [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21]. These data clearly show the association of higher CRP concentrations with poorer outcomes in COVID-19 patients, both in terms of disease severity and mortality. From an analytical point of view, several CRP assays are commercially available [22]. Their results can be considered well harmonized when traceability to WHO 1st International Standard 85/506 is assured through the use of ERM-DA472/IFCC and ERM-DA474/IFCC secondary reference materials as common calibrators [23]. This guarantees, to an extent, the comparability of CRP results, allowing the application to different populations of common decisional cut-offs, when available [24]. However, information about this aspect to permit definition of interpretative criteria for patient classification useful worldwide was often lacking in published literature.

Definition of prognostic value of CRP in COVID-19

In the last few years, hundreds of thousands of papers were published in scientific journals about COVID-19. At the time of writing this article, when performing a search on the PubMed database by using “C-reactive protein” and “COVID-19” as keywords, over 4,200 results were retrieved. While this literal outpour of studies helped fuel the rapid understanding of the disease and the characteristics of affected patients, benefiting the development of clinical management strategies, it is undeniable that, in some cases, the rapidity in publication came to the detriment of paper quality. Particularly in the field of laboratory medicine, many issues already present in medical literature before the pandemic were exacerbated by the sudden influx of publications. It is indeed essential for clinical studies that collection and analysis of laboratory data are rigorously performed to provide robust and replicable results that can be applied universally. This can be achieved by addressing all phases of the total examination process, from preanalytical interferences and sample quality, through the use of standardized analytical methods, to measurement units’ harmonization. Unfortunately, as discussed in previous works [25], [26], [27], [28], [29], [30], these aspects were often overlooked or blatantly absent in a significant portion of literature. This may have led to confusion in result interpretation among different studies and, more importantly, inability of inference of interpretative clinical criteria to other settings wishing to duplicate published experiences.

Because of this confounding situation, in April 2020, i.e., in the early phases of the pandemic, we decided to take the matter in our own hands and perform a study on our hospitalized COVID-19 population for the appraisal of major biochemical predictors of COVID-19 severity, including CRP [31]. The ‘Luigi Sacco’ academic hospital in Milan is one of the two national reference centers for infectious diseases in Italy and, for this reason, in late February 2020 we started to receive patients with COVID-19 from the first epicenters of the SARS-CoV-2 outbreak in northern Italy (and probably in Europe). Between February 21st and March 31st, 2020, 427 COVID-19 patients were hospitalized in our institution, which at that time was probably one of the biggest populations in the world, outside of China. A retrospective study was performed on data collected from these patients for the definition of optimum biomarker cut-offs specifically selected to have a high rule-in ability in detecting patients at risk of in-hospital death (minimizing false positive results) and a high rule-out ability in identifying patients at very low risk of intensive care unit (ICU) admission (minimizing false negative results) [31]. The population of the study was composed of adult subjects with clinical and radiologic findings suggestive of COVID-19 at admission and positive detection of SARS-CoV-2 RNA on nasopharyngeal swab by using a real-time reverse transcription polymerase chain reaction method. The median age was 61 years (interquartile range: 50–73 years) and most patients were male (69 %). 11.0 % of the included patients were admitted to the ICU during hospital stay and 20.8 % died during hospitalization. It is noteworthy that CRP determinations included in the study analysis were performed by using the immunoturbidimetric assay on the Alinity c platform (Abbott Diagnostics) traceable to the ERM-DA472/IFCC reference material, previously shown to assure a good analytical performance for the clinical application of the measurements [24, 32]. For each patient, when more than one CRP result was present, the worst result (i.e., the highest) of the whole hospitalization period was considered for statistical analyses. Table 2 summarizes the findings of the study regarding CRP results in patients, when stratified for different outcomes. The prevalence of elevated CRP (>10 mg/L) in the whole population was 92 %. In terms of prognostic thresholds, the best CRP cut-off found for predicting in-hospital death was >303 mg/L, associated with a specificity of 96 % [95 % confidence interval (CI): 93–98] and a positive predictive value (PPV) of 73 % (CI: 59–85), while CRP concentrations lower than 141 mg/L excluded with a high probability the need of intensive care treatment [sensitivity, 94 %; CI: 83–99; negative predictive value (NPV), 99 %; CI: 94–100] (Figure 1). Out of the six biomarkers evaluated in the study (i.e., serum albumin, lactate dehydrogenase, cardiac troponin T, CRP, D-dimer, and ferritin), CRP had the second highest power (following serum albumin) in predicting ICU admission, with an area under the ROC curve of 0.87 (95 % CI: 0.84–0.93).

Figure 1:

Optimum CRP cut-offs in COVID-19 specifically selected to have a high rule-in ability in detecting patients at risk of in-hospital death (minimizing false positive results) (upper part) and a high rule-out ability in identifying patients at very low risk of intensive care unit (ICU) admission (minimizing false negative results) (lower part). Adapted from ref. [31].

Validation of these CRP cut-offs was later performed on a subsequent independent cohort of 1,604 patients, obtaining similar results in terms of prognostic performance, with a sensitivity and an NPV at CRP <141 mg/L for detecting patients who did not need ICU admission of 86 % (CI: 79–91) and 98 % (CI: 97–99), respectively, and a specificity for predicting in-hospital death of CRP values >303 mg/L equal to 97 % (CI: 95–97). Due to the significant decrease in the number of in-hospital deaths in the two periods of patient enrolment (from 20.8 to 14.7 %), the PPV associated with CRP >303 mg/L decreased however to 57 % (CI: 49–65). Our results generally confirmed data present in literature about the value of CRP testing in COVID-19. However, they had some important peculiarities such as the use of a suitable analytical methodology and the cut-off derivation by using a harmonized assay, permitting to apply them to similar populations worldwide providing that the related institutions also use CRP assays that produce harmonized results. Only the use of assays providing harmonized results allows indeed the use of common decision thresholds, enabling the universal application of results of clinical studies undertaken in different locations or times and permitting their unambiguous interpretation [33].

Impact of CRP measurement uncertainty on prognostic classification

Harmonization of laboratory results for a given test is needed to permit their interpretation in a consistent manner. The definition and implementation of a reference measurement system, based on an unbroken metrological traceability chain linking patients’ results to higher-order references, represents the agreed approach [34]. Thanks to the availability of suitable reference materials, current harmonization of CRP measuring systems appears to be satisfactory [23, 35]. What is still frequently ignored is the entity and the impact of measurement uncertainty (MU) associated with CRP results of clinical samples, including the demonstration of fulfillment of proposed analytical performance specifications (APS), which for CRP standard MU are 3.76 and 5.64 % for desirable and minimum quality of results, respectively [24, 32, 36, 37].

Figure 2 shows a simulation evaluating the impact of different MU associated with clinical results for CRP on the prognostic classification, using the above-mentioned cut-offs (303 mg/L and 141 mg/L), of a cohort of 2045 COVID-19 patients previously recruited for a study focusing on the origin and clinical significance of aspartate aminotransferase increases in these patients [5]. According to the ISO 20914:2019 Technical Specification [38], standard MU associated with CRP measurements of clinical samples can be estimated as √(u_ref² + u_cal² + u_Rw²), where u_ref is the MU associated with the value of the manufacturer’s selected reference material, u_cal is the MU associated with the value of the commercial calibrator, and u_Rw is the random uncertainty of the assay in use in the individual laboratory, obtained under intermediate reproducibility conditions as defined in ISO 20914:2019 [38]. Starting with a standard MU of 4.25 %, corresponding to the current MU performance in the authors’ laboratory [32], and expanding its value with a coverage factor of 2 to apply a 95 % level of confidence to results, an uncertainty interval was built around the measured CRP result of each patient. Then, by considering both the lowest and highest values of that interval compared with the prognostic cut-offs, the rate of possibly misclassified patients was derived for each outcome. This approach was then repeated with gradually increasing expanded MU values. As shown in Figure 2, an expanded MU increase from 8.5 to 11.0 % (although still lower than the minimum APS for expanded MU of 11.3 %) led to an increase in patient misclassification rate for in-hospital death outcome from 4.3 to 5.4 % (Figure 2A) and for unnecessary ICU admission from 8.7 to 11.3 % (Figure 2B), respectively. Even though these increases in patient misclassification rates may appear small in absolute terms, when applied to large populations, it is possible that a significant number of patients may be categorized inaccurately, with the risk of under- or over-treatment and subsequent consequences such as increased length of in-hospital stay, execution of unnecessary procedures, or worsening outcome. Maintaining MU of CRP results as low as possible is therefore crucial for patient care and correct prognostic classification. This is certainly true for COVID-19 patients, as shown in our simulation, but may be even more important when CRP is used as a cardiovascular risk marker, with very low concentration cut-offs, since MU usually has a greater impact on lower measurand concentrations [39].

Figure 2:

Impact of different measurement uncertainties associated with clinical results for CRP on the classification of COVID-19 patients as at high risk of in-hospital death (A) and low risk of intensive care unit (ICU) admission (B). The dashed lines include the absolute number of patients which are potentially misclassified based on gradually increasing uncertainty values.

Conclusions

The well-known role of CRP in managing inflammatory disease has been widely confirmed in COVID-19. Particularly, this test may assist in the prognostic classification of these patients, helping to improve administration of adequate treatments and recognize individuals with severe disease who need additional care. Harmonization of CRP measurements is central to avoid confusion in result interpretation and to allow worldwide application of optimal decision cut-offs. In addition, MU at the level of clinical sample results should be as lower as possible to reduce patient misclassification rates.