Proteinuria Increases the PLASMIC and French Scores Performance to Predict Thrombotic Thrombocytopenic Purpura in Patients With Thrombotic Microangiopathy Syndrome

Introduction PLASMIC and French scores have been developed to help clinicians in the early identification of patients with thrombotic thrombocytopenic purpura (TTP). Nevertheless, the validity of these scores in thrombotic microangiopathy (TMA) cohorts with low TTP prevalence remains uncertain. We aimed to evaluate their diagnostic value in routine clinical practice using an unselected cohort of patients with TMA. We also analyzed the value of adding proteinuria level to the scores. Methods We retrospectively included all patients presenting with a biological TMA syndrome between January 1, 2008, and December 31, 2019, in a tertiary hospital. TMA etiology was ascertained, and scores were evaluated. Modified scores, built by adding 1 point for low proteinuria (<1.2 g/g), were compared with original scores for TTP prediction. Results Among 273 patients presenting with a full biological TMA syndrome, 238 were classified with a TMA diagnosis. Complete scores and proteinuria level were available in 134 patients with a TTP prevalence of 7.5%. Area under the receiver operating characteristic curve (AUC) of PLASMIC and French scores for TTP diagnosis was 0.65 (0.46–0.84) and 0.72 (0.51–0.93), respectively. AUC of modified PLASMIC and French scores was 0.76 (0.59–0.92) (P = 0.003 vs. standard score) and 0.81 (0.67–0.95) (P = 0.069 vs. standard score), respectively. Specificity, positive predictive value (PPV), and positive likelihood ratio of high-risk scores were significantly improved by adding proteinuria level. Conclusion PLASMIC and French scores have low predictive values when applied to an unselected TMA cohort. Including proteinuria level in the original scores improves their performance for TTP prediction.

mild acute renal failure) and laboratory results (TMA syndrome). Given the high rate of early mortality, emergency TTP treatment is mandatory, 5,6 which is based on an association of plasma therapy, corticosteroids, and targeted therapies, including rituximab 7 and caplacizumab. 8,9 Therapeutic plasma exchanges need to be started as soon as diagnosis is suspected. 6 Differential diagnosis between TTP and other TMAs, and especially HUS, remains a challenge to some cases because of an overlap in clinical signs between these conditions. In TMA diagnosis workup, determination of ADAMTS13 activity is the cornerstone; however, the assay remains unavailable in many hospitals. To assist clinician decision-making, several scores have been developed to predict severe ADAMTS13 deficiency, for example, the PLAS-MIC score 10 and the French score. 11 Importantly, although revealing good predictive values, these scores have been developed in TMA cohorts characterized by a high TTP prevalence (29%-63%). These cohorts may not reflect the real prevalence of TTP among patients with TMA. Indeed, TTP prevalence was found to range from 3.2% to 5.6% in the cohorts taking into account all TMA etiologies (unselected TMA cohorts). 12,13 Thus, the predictive value of PLASMIC and French scores still needs to be studied in "real life" TMA cohorts.
In opposition to most other TMA, TTP involves the kidneys less frequently, a characteristic that was historically used to differentiate TTP from HUS. 14 In HUS, the kidneys are the major target of TMA, leading to glomerular microthrombosis and proteinuria. On this basis, it was recently suggested that adding proteinuria level to the French score may enable TTP and HUS to be better distinguished. 15 Thus, the aims of this study were to evaluate the diagnostic value of the PLASMIC and French scores in "real life practice" using an unselected cohort of consecutive patients with TMA and to study the value of the modified scores that include proteinuria level.

Selection of Patients
Adult patients ($18 years old) admitted to the University Hospital of Angers between January 1, 2008, and December 31, 2019, with a full biological TMA syndrome were retrospectively included in the study. A full biological TMA syndrome was defined by the concomitant association of anemia (<12 g/dl in females and 13 g/dl in males), thrombocytopenia (#150 g/l), schistocytosis ($0.5%), and decreased haptoglobin level (#0.4 g/l). Patients were identified from the database of the hematological laboratory. The study protocol complied with the Ethics Committee of the Angers University Hospital (no. 2019/12).

TMA Causes
As described earlier, 13 medical records of patients identified with having a full biological TMA syndrome were first analyzed by 5 physicians trained in nephrology, hematology, and critical care medicine to confirm or rule out a TMA diagnosis. The second step was to identify the etiology of TMA after a hierarchical analysis, according to current classifications 16,17 and as previously described (Supplementary Figure S1). 13 Thus, by using this methodology, we were able to identify a cohort of consecutive patients with TMA with a full biological TMA syndrome and with all etiologies considered, which we term thereafter as "unselected cohort." Data Collection and Score Assessment Demographic, clinical, and biological data at TMA diagnosis were retrospectively retrieved. PLASMIC and French scores were calculated as described 10,11 without considering antinuclear antibodies for the latter (Table 1). Proteinuria was collected on the day of TMA diagnosis. Acute kidney injury was defined using serum creatinine levels and the Kidney Disease Improving Global Outcomes Criteria. 18 ADAMTS13 was collected when available. As previously described, the PLASMIC score was dichotomized into high and low intermediate risk when the score was $6 or #5, respectively. 10 The French score was dichotomized into high and low risk when the score was 2 or #1, respectively. 11 Modified PLASMIC and French Score Assessment Modified PLASMIC and French scores (Table 1) were formulated by adding 1 point when proteinuria level  Figure S2B). Thus, modified PLASMIC and French scores ranged from 0 to 8 and from 0 to 3, respectively. The modified PLASMIC score was dichotomized into high and low intermediate risk when the score was $7 or #6, respectively. The modified French score was dichotomized into high and low risk when the score was equal to 3 or #2, respectively.

Statistical Analysis
Quantitative variables, presented as median (interquartile range), were compared with the Mann-Whitney U test (or Kruskal-Wallis test followed by Dunn post hoc test for multiple comparisons when applicable). Qualitative variables, presented as the absolute value and percentage, were compared using the c 2 test (or Fisher exact test when necessary). TTP diagnosis performance was analyzed using receiver operating characteristic curves. AUCs were compared using a Delong test. 19 Performances of high-risk scores for TTP diagnosis were compared using the McNemar test (for sensitivities and specificities), 20 generalized score statistics (for negative and positive PVs), 21 or a regression model approach (for negative and positive likelihood ratios). 22 Statistical analysis was performed using Prism GraphPad Software version 6.01 (Prism, La Jolla, CA) and R version 4.0. P < 0.05 was considered significant.

Flow Chart of the Study
During the above-mentioned period, we identified 485 patients with thrombocytopenia and schistocytosis ($0.5%) and 5031 patients with haptoglobin (#0.4 g/l). After crossing data sets, we identified 273 patients with a full biological TMA syndrome. After a medical chart review, 238 patients were finally diagnosed with having TMA (28 had no evidence of TMA), in whom 225 patients had all components of the PLASMIC and French scores (thereafter called the "scores cohort"). Proteinuria determination at diagnosis was available in 158 patients. Finally, 134 patients had all components of the PLASMIC score, the French score, and proteinuria determination at diagnosis (thereafter called the "modified scores cohort" [MSC]) ( Figure 1).

Baseline Characteristics of the Cohort
The "scores cohort" (in which TTP prevalence was 6.2%, n ¼ 14) is described in Supplementary Table S1. The "modified scores cohort" (in which TTP prevalence was 7.5%, n ¼ 10) is described in  Table S2). The ability of the PLASMIC score to distinguish TTP from other TMA diagnoses in this population was low Elevated free bilirubin, a n [ Table S2).
The ability of the French score to distinguish between TTP and other TMA diagnoses in this population was in similar ranges (AUC ¼ 0.65 [0.47-0.83], P ¼ 0.06) (Figure 2 and Supplementary Figure S2A). A highrisk (score $ 2) predicted TTP with a sensitivity of 57%, specificity of 81%, PPV of 16%, and NPV of 97%.

Description of Standard and Modified PLASMIC Scores (in MSC)
In MSC, the standard PLASMIC score was similar in patients with TTP (6 [5.5-6.25]) than in all other patients with TMA (5 [5][6], P ¼ 0.08). It was higher in patients with TTP than in patients with a-HUS or drugand transplantation-associated TMA (Table 3).
To build the modified PLASMIC score, we chose to add 1 point (in case of proteinuria level < 1.2 g/g) to the standard scores because it performed better than adding more (Supplementary Figure S3A-C).
The modified PLASMIC score was higher in patients with TTP (7 [5.75-7]) than in all other patients with TMA (6 [5-6], P ¼ 0.004). It was also higher than in  Figure 2. Performance of (a) PLASMIC and (b) French scores to predict TTP. P value refers to the comparison between the standard and modified scores. ADAMTS13, a disintegrin and metalloprotease with thrombospondin type I repeats-13; AUC, area under the receiver operating characteristic curve; HUS, hemolytic uremic syndrome; TTP, thrombotic thrombocytopenic purpura.  Table 3).

Performances of Standard and Modified PLASMIC Scores (in MSC)
In MSC, the AUC of the PLASMIC score was 0.65 ([0.46-0.84], P ¼ 0.12). The AUC of the modified PLASMIC score was 0.76 ([0.59-0.92, P ¼ 0.006]), which was significantly higher than the standard score (P ¼ 0.003) (Figures 2a and 3a In other words, among the 10 patients with TTP diagnosis in MSC, 8 patients (80%) versus 7 patients (70%) were classified in the high-risk group according to the standard or modified PLASMIC score, respectively (P ¼ 1). More importantly, the modified PLASMIC score identified significantly more patients with a TMA of any other etiology (patients with non-TTP) (vs. the standard score): 102 (82%) versus 66 (53%) were in the low intermediate-risk group (P < 0.001) ( Table 3).

Description of Standard and Modified French Scores (in MSC)
The standard French score was higher in patients with TTP (2 [0.75-2]) than in all other patients with TMA (1 , P ¼ 0.01). It was also higher in patients with TTP than in patients with a-HUS or drug-and malignant hypertension-associated TMA (Table 3).
To build the modified French score, we chose to add 1 point (in case of proteinuria level < 1.2 g/g) to the standard scores because, as for the PLASMIC score, it performed better than adding more (Supplementary Figure S3D-F).
The modified French score was higher in patients with TTP (2.5 [1.75-3]) than in all other patients with TMA (1 [1][2], P ¼ 0.0002). It was also higher than in patients with a-HUS or drug-, transplantation-, and malignant hypertension-associated TMA ( Table 3).
When dichotomized into high (score ¼ 3) and low risks, the modified French score (vs. the standard score)  In other words, 7 patients (70%) versus 5 patients (50%) with TTP were classified in the high-risk group according to the standard or modified French score, respectively (P ¼ 0.65). In parallel, 104 patients (84%) versus 119 patients (96%) with non-TTP with a TMA of any other etiology were in the low intermediate-risk group according to the standard or modified French score (P ¼ 0.002) ( Table 3).

Both Standard and Modified Scores Performed Better in Selected Subpopulations
After the exclusion of pregnant women, for whom the cause of TMA is often clear, all these scores, modified or not, were improved (Figure 3b). The addition of proteinuria level tended to improve the standard scores (P ¼ 0.076 and 0.071 for the AUC of the modified vs. standard PLASMIC and French scores, respectively; Figure 2). Again, it improved the specificity and PPV of high-risk patients for TTP diagnosis, with a trend for a better positive likelihood ratio (Table 4, lower panel).
It is worth noting that these modified scores were also better when considering the subpopulation with ADAMTS13 determination (n ¼ 40; Figure 2 and Supplementary Figure S4A) or only TTP and all patients with HUS (n ¼ 21; Figure 2 and Supplementary Figure S4B).

DISCUSSION
Clinicians need to rapidly identify patients with primary TMA and especially those with TTP, for whom therapeutic plasma exchanges or fresh frozen plasma administration must be initiated as soon as possible. Scores have therefore been developed to help identify these patients with TTP. It must be noted that these scores have been evaluated and validated in cohorts with a high TTP prevalence. Nevertheless, as recently reported, these populations do not accurately reflect clinical practice. 12,13 Here, we reveal that PLASMIC and French scores, when applied to an unselected TMA population characterized by a low TTP prevalence, do not effectively identify TTP from other TMA etiologies. Moreover, our results suggest that incorporating proteinuria level into the established scores may improve their predictive value. Bendapudi et al. 10 created the PLASMIC score using 2 cohorts with a high prevalence of TTP (14%-47%) and revealed its accuracy for TTP diagnosis (AUC ¼ 0.91-0.96). Further studies with a high TTP prevalence (25%-70%) confirmed the good diagnostic value of both the PLASMIC and French scores. [23][24][25][26][27][28][29] Nevertheless, they also included highly selected patients recruited among those who had ADAMTS13 measurement. 30 As a consequence, the TTP prevalence was much higher in these study groups than TTP prevalence in clinical practice. Interestingly, a recent metaanalysis evaluated the validity of the PLASMIC score 30 and revealed that a score < 5 was associated with high sensitivity and NPV, suggesting it could be used to rule out TTP and to exclude the need for emergency therapeutic plasma exchanges. This was also the case in our cohort, but at the price of a strong drop in specificity (Supplementary Table S4). In this meta-analysis, which included studies with a median TTP prevalence of 35%, the PPV of the PLASMIC score decreased when TTP prevalence was lower (for a 10% TTP prevalence and PLASMIC score $ 5, the PPV decreased to 21% and the NPV was 100%). In our study, applied to an unselected TMA population with a low TTP prevalence, the standard PLASMIC score was not very successful in predicting TTP.
The diagnostic score proposed by Coppo et al., 11 namely the French score, has the advantage of being more simple to calculate than the PLASMIC score. Using 3 criteria (platelets < 30 g/l, serum creatinine level < 2.273 mg/dl, and, to a lesser extent, positivity for antinuclear antibodies), it was found that the score predicts ADAMTS13 deficiency with a sensitivity of 99%, specificity of 48%, PPV of 85%, and NPV of 93%. 11 Applied to the PLASMIC cohort, the French score had an AUC of 0.88 ([0.83-0.91], P ¼ 0.003) and the high-risk group (presence of 2 criteria) identified 83% of patients with severe ADAMTS13 deficiency. 10 Nevertheless, as observed for the PLASMIC score, when applied to our cohort, the French score was also not very successful in predicting a TTP diagnosis. This observation was predictable as the French score was developed in a TMA cohort that voluntarily excluded patients with secondary TMA. 11 Interestingly, a recent study revealed a decreased sensitivity and specificity of PLASMIC and French scores in older patients (>60 years old) 31 as compared with younger patients. The lower diagnostic value of these scores may be related to less typical presentations of TTP in older patients, as suggested by the authors, but it may also be related to the lower TTP prevalence and enrichment with secondary TMA causes in old patients.
In this study, we reveal that integrating proteinuria level, a marker of renal injury unlikely to be observed in TTP, improved the performance of both the PLASMIC and French scores. When set to $7, the modified highrisk PLASMIC score revealed a significant increase in specificity, PPV, and positive likelihood ratio for TTP diagnosis, at the cost of a slight, but not significant, decrease in sensitivity. In the same way, proteinuria level also improved the performance of a high-risk French score (score ¼ 3) with an increase in specificity, PPV, and positive likelihood ratio for TTP diagnosis when compared with the standard score. In practice, this means that it will be easier to identify patients with non-TTP. Thus, by reducing the false positive rate, the modified scores could be used as part of cost-saving strategies, by reducing unnecessary therapeutic plasma exchanges and inappropriate use of rituximab and/or caplacizumab. 32 Despite not being statistically significant, the loss of sensitivity (and the subsequent falsenegative risk) of the modified scores in relation to the standard scores should be taken into consideration and reminds us that clinical expertise cannot be replaced simply by applying a statistical score. 33 Our study has several limitations, the first of which being its retrospective design. Nevertheless, the low incidence of TTP makes it difficult to carry out prospective studies. Second, we enrolled all patients with data available to calculate the PLASMIC and French scores, but not all patients had an ADAMTS13 assay available, which is the gold standard for TTP diagnosis. Nevertheless, all our patients with TTP had undetectable ADAMTS13 levels and it is known that the mortality of untreated patients with TTP is close to 90%. 34 Thus, it is reasonable to exclude TTP diagnosis in patients for whom no ADAMTS13 assay was performed as we have follow-up data regarding these patients. In addition, the performance of the modified scores with proteinuria level was significantly better in the subpopulation where ADAMTS13 levels were available. Third, we used a French score without antinuclear antibodies because this test was rarely performed in our cohort. Nevertheless, previous studies have revealed that a French score based on both platelet count and serum creatinine has a high predictive value for TTP in selected populations. 10,31 In conclusion, we have revealed that PLASMIC and French scores are not effective tools for assisting in TTP diagnosis when applied to an unselected TMA population with a low TTP prevalence. The inclusion of proteinuria level in the scores may improve their performance, especially specificity, PPV, and positive likelihood ratio.
To confirm our results, testing these modified scores in larger, multicenter cohorts, ideally prospective, of patients with TMA for all of which ADAMTS13 is known, would be of great interest. Then, if all the studies converge to the same results, it would be interesting to consider a prospective study where the initial TMA management would be guided by the results of the modified scores. The modified scores could indeed help clinicians in deciding on the emergency use of TPE and could lead to cost savings. Thus, we believe that these modified scores could have an important place in the algorithm for the management of patients with suspected TTP recently proposed by Coppo et al. 35

DISCLOSURE
All the authors declared no competing interests.

ACKNOWLEDGMENTS
The authors acknowledge Ryan Emerson for language editing.

AUTHOR CONTRIBUTIONS
NF, JFA, and BB contributed to the study conception and design. NF, NH, CM, and MT performed the data collection. NF, NH, CM, JFA, and BB performed the initial data analysis. NF prepared the first draft of the manuscript. NF, JFA, and BB revised the manuscript. All authors read and approved the final manuscript.

SUPPLEMENTARY MATERIAL
Supplementary File (PDF) Figure S1. Hierarchical analysis to identify etiology of thrombotic microangiopathy. Figure S2. Performance of PLASMIC and French scores to predict TTP in SC (A). Determination of the best proteinuria threshold for predicting a TTP diagnosis in MSC (B). Figure S3. Performance of PLASMIC (A-C) and French (D-F) scores in predicting a diagnosis of TTP considering various modified scores. Figure S4. Performance of PLASMIC and French scores in predicting a diagnosis of TTP, considering patients with TTP and HUS (A) and patients with an ADAMTS13 determination available (B). Table S1. Clinical and biological presentation of TMA with PLASMIC and French scores available (in SC, n ¼ 225). Table S2. PLASMIC score, French score, and components according to the different etiologies of TMA (in SC, n ¼ 225). Table S3. Performance of high-risk standard and modified scores to predict TTP according to adjusted or unadjusted threshold. Table S4. Performance of high-risk standard PLASMIC scores with standard ($6) or modified ($5) threshold to predict TTP (in SC, n ¼ 225).