Development and validation of a nomogram for steroid-resistance prediction in immune thrombocytopenia patients

ABSTRACT Objectives Corticosteroid is first-line therapy in immune thrombocytopenia. However, nearly 30% of patients appear in steroid-resistance. Our research analyses the relevant indicators of patients and develops a risk prediction model to predict the poor response to steroid-therapy in ITP patients. Methods We collected data from 111 ITP patients admitted to Xiamen University Zhongshan Hospital from 2013 to 2019 as the training cohort and 65 ITP patients during 2019–2020 as the external validation cohort. Screening significant factors(P < 0.05) in univariate analysis, and further identified to be independent variables in multivariable logistic regression analysis. Incorporated the significant risk factors in and presented them with a nomogram based on independent risk predictors. The nomogram was assessed by receiver operating characteristics curves and decision curve analysis. Results We constructed a steroid-resistance prediction model based on the potential predictors including age, serum ferritin and expression of HBsAg. As a result, based on the area under the ROC curves, the training cohort (AUC: 0.718, 95% CI: 0.615–0.821) and the external validation cohort (AUC:0.799,95%CI:0.692–0.905), which displayed good discrimination. The decision curve showed that predicting the steroid-refractory risk in ITP patients using this nomogram with a range of the threshold probability between >16% and <70%. The nomogram appears good performance in predicting steroid-refractory ITP patients. Conclusion Prediction model shows that elder patients with a high level of ferritin and positive expression of HBsAg may appear a high possibility of steroid-resistance. For these patients, TPO-RAs can be considered to help patients to get better treatment effects and develop a better health-related quality of life.


Introduction
Immune thrombocytopenia (ITP) is characterized by immune-mediated peripheral platelet destruction and impaired platelet production with a consequent increased risk of bleeding. Recommended first-line treatment of symptomatic ITP is glucocorticoids, but the majority of adult patients relapse within the first year of treatment and require additional therapy [1][2][3][4][5]. A study shows that 98% of patients with corticosteroid exposure experienced one or more side events, and 38% of patients need to stop or reduce corticosteroid therapy [6]. Biomarkers are needed to identify steroidresistant ITP patients before prolonged courses of systemic steroids, which can lead to serious adverse events and worsen the health-related quality of life (HRQoL), and allow the early introduction of alternative therapy before bleeding symptoms occurs. This research aims to develop a new prediction model to evaluate whether ITP patients are at highrisk of corticosteroid resistance, and help clinicians to choose better therapy.

Patients
We comprehensively collected data from 111 ITP patients admitted to Xiamen University Zhongshan Hospital from May 2013 to May 2019 as a training cohort and 65 ITP patients during June 2019 to June 2020 as an external validation cohort. All patients met the Consensus of Chinese experts on the diagnosis and treatment of adult primary immune thrombocytopenia (version 2016). Baseline clinicopathologic data, including age, gender, serum ferritin(SF), C-creative protein(CRP), erythrocyte sedimentation rate(ESR), procalcitonin(PCT), albumin, globulin, immunoglobulin G (IgG), immunoglobulin M(IgM), immunoglobulin A (IgA), rheumatic related antibody, Epstein-Barr Virus deoxyribonucleic acid(EBV-DNA), Hepatitis B surface antigen(HBsAg). Patients received standard corticosteroids therapy doses (eg, prednisone1 mg/kg/d, dexamethasone 40 mg/d×4) then gradually reduced the doses. For 3-6 months treatment, the patients still needed a large dose (prednisone 15 mg/d) or more to keep platelet count >30 × 10 9 /L or no response were defined as the steroid-resistance group and others as the non-steroid-resistance group. We defined rheumatism related antibody such as the positive expression of anti-Sjögren's syndrome A/B antibody (SSA/B) and so on while cannot meet the diagnostic criteria for rheumatism related diseases as the positive expression of rheumatism related antibody. Patients with positive expression of HBsAg, and the absence of HBeAg and presence of anti-HBe, undetectable or low levels of HBV DNA in PCR-based assays, with normal ALT levels called the inactive HBsAg carrier state [7][8][9]. Only hepatitis B virus carriers are enrolled in our study, and the Hepatic dysfunction HBV patients had excluded.
This research protocol was conducted following the Declaration of Helsinki and had been approved by the Ethics Committee of Xiamen University Zhongshan Hospital. We reviewed and collected the relevant medical records and follow-up data after the patient's informed consent.

Statistical analysis
All of the data were analyzed using R software (version 3.6.3; https://www.R-project.org). The significance level for all of the statistical tests was set at 0.05. All statistical significance levels were two-sided.
Patient characteristics and factors were analyzed using student's t-test and chi-square tests. Ages are given as the medians with ranges, other variables are expressed as count (%), and performed univariate analysis to examine the difference between the steroid-resistance group and the non-steroid-resistance group. Apply binary logistic regression modeling technique to analyze risk factors for steroid-refractory ITP patients. Variables that had a p-value of < 0.05 in univariate analysis were selected into the multivariable logistic analysis to get further independent risk factors. In the multivariable logistic analysis, variables with a pvalue of < 0.05 were identified to be the independent risk factors automatically and finally selected into the final model. Regression coefficients were used to generate prognostic nomograms. Model discrimination was measured quantitatively with the concordance index. Internal validation was performed using 1000 bootstrap resampling to quantify the overfitting of our modeling strategy and predict the future performance of the model. 65 new patients we collected were used as the external validation cohort to evaluate the model's practicality.

Patient characteristics
We analyzed 176 ITP patients admitted to Xiamen University Zhongshan Hospital from 2013 to 2020 and we divided the patients into a training cohort (n1 = 111) and an external validation cohort (n2 = 65) ( Figure 1). The characteristics of patients in the training cohort and the external validation cohort are summarized in Table 1.

Predictors for steroid-refractory ITP patients
A total of 43 in 111 ITP cases of training cohort appeared steroid-resistance (38.7%). The univariate logistic regression analyses showed eight variables (P < 0.05) were significantly correlated with steroidrefractory ITP patients (Table 2) including age, CRP, globulin, IgG, IgM, IgA, serum ferritin and HBsAg.
To explore the independent risk factors for steroidrefractory ITP patients, we performed a multivariable analysis based on the results of the univariate analysis. Eight risk factors aforementioned were included in the logistic regression model. Finally, age (odds ratio, OR 0.973, 95% CI, 0.951-0.995), high level of serum ferritin (OR 2.570, 95% CI,1.051-6.286), and the positive expression of HBsAg (OR 3.482, 95%CI,1.492-8.126) were identified as risk factors compared to nonsteroid-refractory ITP patients (Table 3).

Development of the prediction model
The above three independent predictors of steroidrefractory ITP patients were integrated into a steroidrefractory ITP patients risk estimation nomogram ( Figure 2).

Apparent performance and validation of the nomogram
The calibration curves of the nomogram for predicting steroid-resistance in ITP patients appeared in good agreement in the training and the external validation cohort (Figure 3). We constructed receiver operating characteristic curves (ROC) for the prediction model including the training and the external validation cohort ( Figure 4). Based on the area under ROC curves (AUC), the training cohort (AUC: 0.718, 95% CI: 0.615-0.821) and the external validation cohort (AUC:0.799, 95%CI:0.692-0.905), which suggested the model has a good discrimination and prediction capability in predicting the steroid-resistance in ITP patients.

Clinical use
Decision Curve Analysis (DCA) was performed in Figure  5, which showed that predicting the steroid-refractory risk in ITP patients using this nomogram would be better than other patients with a range of the threshold about between >16% and <70%.

Discussion
Our research found that age, high level of serum ferritin and the positive expression of HBsAg were three Besides iron overload, frequent blood transfusion, hepatocellular disease and malignant tumor, ferritin elevation can be reactive, especially in inflammatory disease, which indicates the activation of macrophages [10,11]. Acute myeloid leukemia (AML) patients with a high level of serum ferritin overall survival(OS) and recurrence-free survival(RFS) were significantly reduced [12]. Patients with a high level of serum   ferritin in nasopharyngeal carcinoma (NPC), multiple sclerosis (MS), acute exacerbation of idiopathic pulmonary fibrosis (AE-IPF), and many liver diseases tend to suffer disease progression and develop a poor prognosis [13][14][15][16][17]. As for treatment, series of studies show that in liver transplantation, adult-onset still's disease (AOSD), macrophage activation syndrome (MAS) and acute lupus hemophagocytic syndrome (ALHS), patients with a high level of serum ferritin have a high mortality rate and a poor response to corticosteroid therapy [16,[18][19][20][21][22]. Eltrombopag, a TPO receptor agonist, can promote the proliferation of bone marrow megakaryocytes and increase megakaryocytes to form platelets [23]. Moreover, eltrombopag is also a powerful iron chelator, which decreases the total cellular iron and ferritin in different cell lines [24]. A study of small samples found that a majority of children ITP patients appeared to therapy-induced iron deficiency with the treatment of eltrombopag [25]. Studies pointed out that eltrombopag and dexamethasone have synergism, which reduces the level of serum ferritin while raising platelets [24,26,27]. In our center, some ITP patients with a high level of serum ferritin get a longer period of remission and decrement of serum ferritin for the treatment of eltrombopag combined with high-dose dexamethasone. This requires further prospective study to verify.  Although most ITP patients are primary, hepatitis viruses can cause ITP [28][29][30][31], including hepatitis A, hepatitis B, hepatitis C, hepatitis E, which can cause immune dysfunction and affects the function of megakaryocyte [32][33][34][35][36][37]. Series of studies indicated that the sensitivity of HCV-ITP patients to corticosteroids was significantly lower than the one without HCV and was correlated with HCV level [38][39][40]. Different from foreign countries, the most common form of viral hepatitis in China is hepatitis B. Our study found that ITP patients with positive expression of HBsAg appear poor responses to corticosteroid. There are occult hepatitis B infection(OBI) cases that occur the reactivation of hepatitis B virus and lead to hepatic dysfunction when using rituximab, corticosteroid and other immunity inhibitors [28,41,42]. The treatment of ITP patients with hepatitis B should be specific but there is no consensus so far. Studies pointed out that eltrombopag can increase the counts of platelets and had been approved for the treatment of hepatitis C virus-associated immune thrombocytopenia(HCV-ITP) [43][44][45]. However, there are few studies about the therapy in HBV infected ITP patients, and require further discussion.
Our research found that age was a risk factor for steroid-resistance. For elder patients, glucocorticoid therapy, rituximab and other immune suppressive drugs tend to have more side-effects [46][47][48]. As for splenectomy therapy may grow further immunosuppressive and increases the chance of severe infection [46,49,50]. In this case, eltrombopag may become a safe and effective choice [51][52][53].

Conclusions
Our prediction model shows that elder patients with a high level of ferritin and positive expression of HBsAg may appear a high possibility of steroid-resistance. It can help clinicians identify steroid-refractory ITP patients early. For these patients, TPO-RAs combined with first-line corticosteroid therapy can be considered to help patients to get better treatment effects and develop a better health-related quality of life.  . Decision curve analysis for steroid-refractory ITP patients. The y-axis represents the net benefit, and the xaxis represents the steroid-refractory threshold probability in ITP patients. The blue line represents the prediction nomogram of steroid-refractory ITP patients. The gray line represents the assumption that all patients were steroidresistant. The black line represents the assumption that no patients were steroid-resistant. The decision curve showed that predicting the steroid-refractory risk in ITP patients using this nomogram with a range of the threshold probability between >16% and <70%.