本研究旨在为临床医生开发一种定量方法，以预测新型冠状病毒肺炎（COVID-19）患者临床预后好转的可能性。本研究收集了2020年1月10日至2月26日入院后经实验室确诊的104例COVID-19感染患者的数据，包括患者的临床基本信息和实验室检查结果，并比较好转和未好转患者的各项参数。本研究使用最小绝对收缩和选择算法（LASSO）logistics 回归模型和双向逐步策略的多因素logistics 回归模型筛选预测预后因素，使用一致性指数（C指数）评估模型，并通过重复抽样进行内部验证，以此构建了一项新的预测列线图。截至2020年2月26日，研究中包括的104位患者（中位年龄为55岁）中，75位（72.1%）预后好转，而29位（27.9%）没有明显好转迹象。临床预后好转的患者与未好转的患者在临床特征和实验室检查结果上存在许多差异。经过多步筛选过程后，本研究筛选出5项预后因素并将其纳入列线图的构建，包括免疫球蛋白A（IgA）、C反应蛋白（CRP）、肌酸激酶（CK）、急性生理学和慢性健康评估表II（APACHE II），以及CK和APACHE II之间的相互作用。本研究建立的模型的C指数为0.962 [95%置信区间（CI）为0.931~0.993]，并且通过重复抽样验证其值仍然达到0.948。预测列线图与理想模型相比，在校准图方面显示出接近的性能，并且决策曲线和临床影响曲线显示，其在临床上具有实用性。本研究构建的列线图有助于临床医生预测每位COVID-19患者的临床预后好转的可能性，将有助于个性化的咨询和治疗。

The aim of this research was to develop a quantitative method for clinicians to predict the probability of improved prognosis in patients with coronavirus disease 2019 (COVID-19). Data on 104 patients admitted to hospital with laboratory-confirmed COVID-19 infection from 10 January 2020 to 26 February 2020 were collected. Clinical information and laboratory findings were collected and compared between the outcomes of improved patients and non-improved patients. The least absolute shrinkage and selection operator (LASSO) logistics regression model and two-way stepwise strategy in the multivariate logistics regression model were used to select prognostic factors for predicting clinical outcomes in COVID-19 patients. The concordance index (C-index) was used to assess the discrimination of the model, and internal validation was performed through bootstrap resampling. A novel predictive nomogram was constructed by incorporating these features. Of the 104 patients included in the study (median age 55 years), 75 (72.1%) had improved short-term outcomes, while 29 (27.9%) showed no signs of improvement. There were numerous differences in clinical characteristics and laboratory findings between patients with improved outcomes and patients without improved outcomes. After a multi-step screening process, prognostic factors were selected and incorporated into the nomogram construction, including immunoglobulin A (IgA), C-reactive protein (CRP), creatine kinase (CK), Acute Physiology and Chronic Health Evaluation II (APACHE II), and interaction between CK and APACHE II. The C-index of our model was 0.962 (95% confidence interval (CI), 0.931–0.993) and still reached a high value of 0.948 through bootstrapping validation. A predictive nomogram we further established showed close performance compared with the ideal model on the calibration plot and was clinically practical according to the decision curve and clinical impact curve. The nomogram we constructed is useful for clinicians to predict improved clinical outcome probability for each COVID-19 patient, which may facilitate personalized counselling and treatment.