MO1053FACTORS INFLUENCING DROPOUT FROM PHYSICAL FUNCTION ASSESSMENT PROGRAMS AMONG PATIENTS RECEIVING MAINTENANCE HEMODIALYSIS

Abstract Background and Aims Continuing participation in physical function assessment programs is a critical component of treatment for patients undergoing hemodialysis (HD). Maintaining physical function through participation in a physical function assessment program is important to prevent adverse events in the clinical field of nephrology. Several clinical practice guidelines recommend regular assessment of physical activity and physical function as part of routine care for patients receiving hemodialysis (HD). However, the factors related to patients’ continuation of a physical function assessment program while undergoing HD remain unknown. We aimed to investigate the predictors associated with dropout from a physical function assessment program among patients receiving outpatient HD. Method In 2016, Seirei Sakura Citizen Hospital initiated a physical function assessment program for patients receiving outpatient HD. This retrospective cohort study included 230 patients receiving HD who participated in the first physical function assessment program in 2016. Following the initial visit, all patients were invited to complete a physical function assessment once a year. We assessed self-efficacy (SE), short physical performance battery (SPPB), exercise habits, and hand grip and provided patients with appropriate feedback. These measures were performed before the hemodialysis session. Laboratory Data and dialysis status were also collected. Participants were tracked for three years after their first physical function assessment to determine their attendance rate. Patients were provided with four opportunities for participation, including the initial assessment. The program's participation rate was defined as the number of program sessions in which the patient actually participated and the percentage (%) of the four physical functioning assessment visits attended. Patients were then divided into a continuation group (> 50% participation, including the initial assessment) and a dropout group (≤ 50% participation, including the initial assessment). Multivariate logistic regression analyses were used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) for the continuation group to determine the predictors of dropout from the physical function assessment program. Ethical approval was provided by Seirei Sakura Citizen Hospital,and written informed consent. Results A total of 230 patients receiving outpatient HD were invited to participate in the study. Among these, 78 patients refused to participate, 45 participants died or changed clinic within three years of obtaining baseline measurements, and six patients had missing data. Therefore, the final analysis included 101 patients undergoing HD. The continuation and dropout groups included 43 and 58 patients, respectively. SE (continuation: 13.0 ± 4.3 points; dropout: 9.8 ± 4.8 points) and age (continuation: 65.7 ± 10.4 years; dropout: 61.2 ± 12.2 years) were significantly higher in the continuation group than in the dropout group (p = 0.001, p = 0.047, respectively). Multivariate logistic regression analyses indicated that only SE (OR: 1.192, 95% CI: 1.088–1.319) remained a significant predictor after adjustment (p < 0.05). Conclusion Our data demonstrate that exercise-related SE and age significantly influenced dropout from the physical function assessment program; in particular, SE was a strong predictor of dropout, possibly because the patients with high SE may have had positive feelings about exercise based on previous experience. The older patients may have had a smaller social circle and more time to spare and been more aware of their health and desires. There is a need to evaluate SE to prevent dropout from physical functioning assessment programs. Interventions designed to enhance exercise-related SE may improve program retention among patients with HD.

Download Full-text

Predictors of thermal increase in magnetic resonance–guided focused ultrasound treatment for essential tremor: histogram analysis of skull density ratio values for 1024 elements

Journal of Neurosurgery ◽

10.3171/2021.5.jns21669 ◽

2021 ◽

pp. 1-6

Author(s):

Ken Iijima ◽

Hajime Yokota ◽

Toshio Yamaguchi ◽

Masayuki Nakano ◽

Takahiro Ouchi ◽

...

Keyword(s):

Logistic Regression ◽

Operating Characteristic ◽

Focused Ultrasound ◽

Characteristic Curve ◽

Density Ratio ◽

Predictive Ability ◽

Curve Analysis ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Operating Characteristic Curve

OBJECTIVE Sufficient thermal increase capable of generating thermocoagulation is indispensable for an effective clinical outcome in patients undergoing magnetic resonance–guided focused ultrasound (MRgFUS). The skull density ratio (SDR) is one of the most dominant predictors of thermal increase prior to treatment. However, users currently rely only on the average SDR value (SDRmean) as a screening criterion, although some patients with low SDRmean values can achieve sufficient thermal increase. The present study aimed to examine the numerical distribution of SDR values across 1024 elements to identify more precise predictors of thermal increase during MRgFUS. METHODS The authors retrospectively analyzed the correlations between the skull parameters and the maximum temperature achieved during unilateral ventral intermediate nucleus thalamotomy with MRgFUS in a cohort of 55 patients. In addition, the numerical distribution of SDR values was quantified across 1024 elements by using the skewness, kurtosis, entropy, and uniformity of the SDR histogram. Next, the authors evaluated the correlation between the aforementioned indices and a peak temperature > 55°C by using univariate and multivariate logistic regression analyses. Receiver operating characteristic curve analysis was performed to compare the predictive ability of the indices. The diagnostic performance of significant factors was also assessed. RESULTS The SDR skewness (SDRskewness) was identified as a significant predictor of thermal increase in the univariate and multivariate logistic regression analyses (p < 0.001, p = 0.013). Moreover, the receiver operating characteristic curve analysis indicated that the SDRskewness exhibited a better predictive ability than the SDRmean, with area under the curve values of 0.847 and 0.784, respectively. CONCLUSIONS The SDRskewness is a more accurate predictor of thermal increase than the conventional SDRmean. The authors suggest setting the SDRskewness cutoff value to 0.68. SDRskewness may allow for the inclusion of treatable patients with essential tremor who would have been screened out based on the SDRmean exclusion criterion.

Download Full-text

Middle Cerebral Artery-to-Uterine Artery Pulsatility Index Ratio Independently Predicts Adverse Perinatal Outcome in Pregnancies at Term

10.21203/rs.3.rs-51822/v1 ◽

2020 ◽

Author(s):

Sufen Zhou ◽

Hongyan Guo ◽

Heng Liu ◽

Mingqun Li

Keyword(s):

Risk Factors ◽

Logistic Regression ◽

Middle Cerebral Artery ◽

Perinatal Outcome ◽

Pulsatility Index ◽

Uterine Artery ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Adverse Perinatal Outcome ◽

Independent Risk Factors

Abstract Background: This study aimed to investigate potential predictors, including cerebroplacental ratio (CPR), middle cerebral artery (MCA)/uterine artery pulsatility index (PI) ratio, for adverse perinatal outcome in pregnancies at term.Methods: This was an observational, prospective study of recruited pregnancies at term. An adverse perinatal outcome was set as the primary observational endpoint. The receiver operating characteristic (ROC) curve was plotted to investigate the predictive and cut-off values of risk factors for adverse perinatal outcome. Independent risk factors (maternal, neonatal, prenatal ultrasound and Doppler variables) for adverse perinatal outcome were evaluated by the univariate and multivariate logistic regression analyses.Results: A total of 392 pregnancies at term were included and 19.4% of them had suffered adverse perinatal outcome. CPR (OR: 0.42, 95%CI: 0.20-0.93, P=0.032) and MCA/uterine artery PI ratio (OR: 0.25, 95%CI: 0.16-0.42, P=0.032) were two independent risk factors for adverse perinatal outcome by univariate and multivariate logistic regression analyses.Conclusions: MCA/uterine artery PI ratio is a good predictor of adverse perinatal outcome in pregnancies at term.

Download Full-text

Predictive value of serum procalcitonin for both initial and repeated immunoglobulin resistance in Kawasaki disease: a prospective cohort study

Pediatric Rheumatology ◽

10.1186/s12969-019-0379-5 ◽

2019 ◽

Vol 17 (1) ◽

Cited By ~ 2

Author(s):

Shuran Shao ◽

Chunyan Luo ◽

Kaiyu Zhou ◽

Yimin Hua ◽

Mei Wu ◽

...

Keyword(s):

Logistic Regression ◽

Regression Analysis ◽

Kawasaki Disease ◽

Logistic Regression Analysis ◽

Predictive Value ◽

Laboratory Data ◽

Multivariate Logistic Regression Analysis ◽

Multivariate Logistic Regression ◽

Ivig Resistance ◽

Resistance Prediction

Abstract Background Intravenous immunoglobulin (IVIG) resistance prediction is one pivotal topic of interests in Kawasaki disease (KD) since those patients with KD resistant to IVIG might improve of an early-intensified therapy. Data regarding predictive value of procalcitonin (PCT) for IVIG resistance, particularly for repeated IVIG resistance in KD was limited. This study aimed to testify the predictive validity of PCT for both initial and repeated IVIG resistance in KD. Methods A total of 530 KD patients were prospectively recruited between January 2015 and March 2019. The clinical and laboratory data were compared between IVIG-responsive and IVIG-resistant groups. Multivariate logistic regression analysis was applied to determine the association between PCT and IVIG resistance. Receiver operating characteristic (ROC) curves analysis was further performed to assess the validity of PCT in predicting both initial and repeated IVIG resistance. Results The serum PCT level was significantly higher in initial IVIG-resistance group compared with IVIG-response group (p = 0.009), as well as between repeated IVIG responders and nonresponders (p = 0.017). The best PCT cutoff value for initial and repeated IVIG resistance prediction was 1.48 ng/ml and 2.88 ng/ml, respectively. The corresponding sensitivity was 53.9 and 51.4%, while the specificity were 71.8 and 73.2%, respectively. Multivariate logistic regression analysis failed to identify serum PCT level as an independent predictive factor for both initial and repeated IVIG resistance in KD. Conclusions Serum PCT levels were significantly higher in IVIG nonresponders, but PCT may not be suitable as a single marker to accurately predict both initial and repeated IVIG resistance in KD.

Download Full-text

Urinary neutrophil gelatinase-associated lipocalin values alone and combined with Prognostic Index predict septic AKI, DIC, and shock: a pilot study

BMC Research Notes ◽

10.1186/s13104-020-05232-w ◽

2020 ◽

Vol 13 (1) ◽

Author(s):

Yuichiro Shimoyama ◽

Osamu Umegaki ◽

Noriko Kadono ◽

Toshiaki Minami

Keyword(s):

Logistic Regression ◽

Independent Predictor ◽

Prognostic Index ◽

Prognostic Scores ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Significant Independent Predictor ◽

Neutrophil Gelatinase Associated Lipocalin ◽

Life Threatening ◽

Neutrophil Gelatinase

Abstract Objective Sepsis is a syndrome involving life-threatening organ dysfunction. The present study aimed to determine whether septic AKI, ARDS, DIC, and shock can be predicted more readily by combining uNGAL values and inflammation-based prognostic scores, over the use of uNGAL values alone. Results ROC curve analyses yielded the following cut-off values: AKI: 438.5 (ng/ml) for uNGAL at Day 1 (AUC, 0.8), 476.9 (ng/ml) for uNGAL at Day 2 (AUC, 0.86), 123.8 (ng/ml) for uNGAL at Day 3 (AUC, 0.81), 133.6 (ng/ml) for uNGAL at Day 4 (AUC, 0.78), 1.0 for iNS NGAL-NLR (AUC, 0.75), 2.0 for iNS NGAL-PI (AUC, 0.77), DIC; 648.5 (ng/ml) for uNGAL at Day 1 (AUC, 0.77); shock; 123.8 (ng/ml) for uNGAL at Day 3 (AUC, 0.71) and 9 for SOFA (AUC, 0.71). Multivariate logistic regression analyses revealed iNS NGAL-PI to be a significant independent predictor of AKI (OR, 20.62; 95% CI, 1.03–412.3; p = 0.048).

Download Full-text

Genetic polymorphisms associated with clostridium difficile infection in multiple myeloma patients undergoing autologous stem cell transplantation.

Journal of Clinical Oncology ◽

10.1200/jco.2020.38.15_suppl.8522 ◽

2020 ◽

Vol 38 (15_suppl) ◽

pp. 8522-8522

Author(s):

Issam Hamadeh ◽

Zainab Shahid ◽

Manisha Bhutani ◽

Jai Narendra Patel ◽

Nury Steuerwald ◽

...

Keyword(s):

Logistic Regression ◽

Stem Cell ◽

Antibiotic Prophylaxis ◽

Conditioning Regimen ◽

High Dose ◽

Host Immune System ◽

Cell Transplant ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Buccal Swabs

8522 Background: CDI is the primary cause of infectious diarrhea in immunocompromised patients including those undergoing autologous stem cell transplant (SCT). Given the key role of gut microbiome and its interaction with host immune system, we investigated whether polymorphisms in innate immunity genes (identified through Ingenuity Pathway Analysis) were associated with CDI. Methods: We queried our database to identify MM patients who underwent an autologous SCT between April 2015-June 2019. Patients who had their buccal swabs collected through an IRB approved specimen collection protocol were included herein. Data were collected on age, conditioning regimen, CDI diagnosis, time from admission until CDI diagnosis, absolute neutrophil count (ANC) at time of CDI diagnosis, and antibiotic prophylaxis. Genomic DNA was extracted from buccal swabs and genotyped for 62 single nucleotide polymorphisms (SNPs) in ASPH , RLBP1L1, ATP7B, IL-8, FAK, TNFRSF14, CTH, TLR and IL-4. Univariate and multivariate logistic regression analyses were performed to assess association between CDI and presence of SNPs in these genes. Results: A total of 83 patients were identified (25 cases and 58 controls). Baseline characteristics were comparable between two groups. Median age was 67 years (range: 50-79). All patients received high dose melphalan as conditioning, and the same antibiotic prophylaxis during peri-transplant period. Median time from hospitalization until CDI diagnosis was 10 days (IQR:9 days), and median ANC was 0.7/mL (IQR:1.6/mL). Two SNPs (rs2227307 T > G in IL-8 and rs2234167 G > A in TNFRSF14) were significantly associated with CDI risk in both univariate and multivariate logistic regression analyses (Table). Conclusions: Our findings suggest that rs227307G (in IL-8) and rs2234167A (in TNFRSF14) alleles are potential risk factors for CDI after autologous SCT. Our findings, if validated in a larger cohort, would support genetic testing as a screening tool to identify patients who might benefit from prophylaxis against CDI. [Table: see text]

Download Full-text

Musculoskeletal Problems in Relation to Farming, Gender and Smoking

Proceedings of the Human Factors and Ergonomics Society Annual Meeting ◽

10.1177/154193120004403052 ◽

2000 ◽

Vol 44 (30) ◽

pp. 5-588-5-590

Author(s):

M G Björkstén ◽

A. Rask-Andersen

Keyword(s):

Logistic Regression ◽

General Population ◽

Smoking Habits ◽

Control Group ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Male And Female ◽

Musculoskeletal Problems ◽

Female Farmers ◽

Group Belonging

The aim of the present study was to investigate if smoking habits covariated with musculoskeletal problems among a group of male and female farmers and a control group from the general population. A questionnaire was sent to all farms and to a group of controls from the general population in the county of Uppsala. It comprised questions about smoking habits and musculoskeletal problems. Crosstabulations and multivariate logistic regression analyses were performed. In the analyses we included smoking habits, age, gender and group belonging, e g farmers or controls. Age did not give a higher risk for musculoskeletal problems in any of the groups. The results showed that problems were related both to gender and group combined or not combined with smoking habits.

Download Full-text

The association between serum uric acid and controlled attenuation parameter in United States adults without hepatitis B and C: a cross-sectional study

10.21203/rs.3.rs-439398/v1 ◽

2021 ◽

Author(s):

Mingming He ◽

Lihong Chen ◽

Yu Wang ◽

Haijun Ma

Keyword(s):

United States ◽

Logistic Regression ◽

Uric Acid ◽

Hepatitis B ◽

Positive Association ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Controlled Attenuation Parameter ◽

Cross Sectional ◽

The Relationship

Abstract Background: Controlled attenuation parameter (CAP) is a kind of widespread popular parameter to evaluate various types of hepatic steatosis by liver ultrasound transient elastography. We investigated the relationship between serum uric acid (SUA) and CAP without hepatitis B and C virus-infected in the United States adults, data from National Health and Nutrition Examination Survey (NHANES).Methods: The present study was cross-sectional research. 4319 American men and women participants ≥18 years old, without B and C hepatitis, were included in our analysis. There are some measures to evaluate the association between SUA and CAP by multivariate logistic regression analyses, fit smoothing curves, generalized additive models, two-piecewise linear regression model and subgroup analyses.Results: There was a positive association between the value of SUA and CAP by multivariate logistic regression analyses after adjusting for various confounders. Besides, the inflection point of non-linear curve relationship was identified as 4.3 mg/dL, for SUA <4.3 mg/dL, the effect size is 10.6 (P< 0.01); Correspondingly, SUA≥ 4.3 mg/dL, the effect size is 4.3 (P< 0.01).Finally, SUA was positively associated with glycohemoglobin less than 6.5% individuals (β =7.3, P< 0.01) and Fasting glucose less than 7.0 mmol/L individuals (β = 6.8, P< 0.01) in the subgroup analysis. Conclusions: Our research found the relationship between SUA and CAP is non-linear. Subgroup analysis indicated that the positive association between SUA and CAP were showed in non-diabetic patients but not in diabetic.

Download Full-text

Incidence and risk factors for hypoglycemia in patients with hepatocellular carcinoma

10.21203/rs.3.rs-905429/v1 ◽

2021 ◽

Author(s):

XiaoJing Zheng ◽

Hong-Hong Yan ◽

Bin Gan ◽

Xiao-Ting Qiu ◽

Jie Qiu ◽

...

Keyword(s):

Risk Factors ◽

Hepatocellular Carcinoma ◽

Logistic Regression ◽

Hba1c Level ◽

Glycated Hemoglobin ◽

Multivariate Logistic Regression Analysis ◽

Cancer Center ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Factors Associated

Abstract AimTo evaluate the incidence and risk factors for hypoglycemia in patients with hepatocellular carcinoma (HCC).MethodsWe collected and analyzed the clinical data of patients with HCC in our cancer center between April 2020 and June 2021. Univariate and multivariate logistic regression analyses were performed to identify the risk factors associated with hypoglycemia.ResultsThe incidence rate of hypoglycemia in patients with HCC was 28.9% (67/232). Multivariate logistic regression analysis showed a significant association between hypoglycemia and Child-Pugh grade C (odds ratio [OR]=7.3, 95% confidence interval [CI] 2.28–23.31, p=0.001), alpha-fetoprotein (AFP) level (OR=1.000035, 95% CI 1.000007–1.000063, p=0.015), and glycated hemoglobin (HbA1c) level (OR=0.46, 95% CI 0.29–0.73, p=0.001).ConclusionChild-Pugh stage and HbA1c and AFP levels were associated with hypoglycemia in patients with HCC. Our study suggests that these three factors should be comprehensively considered when estimating the risk of hypoglycemia in these patients, and the diagnosis, treatment, and nursing plan should be adjusted in time to reduce the incidence of hypoglycemia.

Download Full-text

Clinical and Genomic Factors Are Predictive of Response and Prognostic of Progression-Free Survival in Patients with Waldenström Macroglobulinemia Treated with Ibrutinib

Blood ◽

10.1182/blood-2019-125069 ◽

2019 ◽

Vol 134 (Supplement_1) ◽

pp. 2823-2823

Author(s):

Jorge J. Castillo ◽

Joshua Gustine ◽

Maria Demos ◽

Andrew Keezer ◽

Kirsten Meid ◽

...

Keyword(s):

Logistic Regression ◽

Serum Albumin ◽

Research Funding ◽

Regression Models ◽

Progression Free Survival ◽

Regression Analyses ◽

Multivariate Logistic Regression ◽

Free Survival ◽

Lower Odds

Introduction: The Bruton tyrosine kinase inhibitor ibrutinib is the only FDA approved therapy for the treatment of symptomatic Waldenstrom macroglobulinemia (WM), and has been associated with high response rates and durable progression-free survival (PFS). Factors associated with depth of response and PFS duration are not well established. We performed a retrospective study aimed at identifying predictive and prognostic factors in WM patients treated with ibrutinib. Methods: We included consecutive patients with a diagnosis of WM treated with ibrutinib monotherapy evaluated at the Dana-Farber Cancer Institute since January 2012 through March 2019. Patients with Bing-Neel syndrome (WM involving the central nervous system) were excluded. Baseline clinical and laboratory characteristics were gathered. MYD88 and CXCR4 mutations were assessed using polymerase chain reaction assays and Sanger sequencing. Responses at 6 months were assessed using criteria from IWWM3. PFS was defined as the time from ibrutinib initiation until last follow-up, death or progression. Univariate and multivariate logistic regression models were fitted for partial response (PR) and very good partial response (VGPR) at 6 months, and Cox proportional-hazard regression models were fitted for PFS. Results: A total of 252 patients were included in our analysis. Selected baseline characteristics include: age ≥65 years (60%), hemoglobin <11.5 g/dl (68%), platelet count <100 K/uL (12%), albumin <3.5 g/dl (39%), b2-microglobulin ≥3 mg/l (70%), serum IgM level ≥7,000 mg/dl (6%), bone marrow involvement ≥60% (54%), previously untreated for WM (33%), time to ibrutinib <3 years (46%). MYD88 L265P and CXCR4 mutations were detected in 98% and 38% of patients, respectively. At 6 months, 71% of patients obtained PR, and 17% VGPR. Multivariate logistic regression analyses showed higher odds of PR at 6 months for hemoglobin <11.5 g/dl (78% vs. 56%; OR 2.8, 95% CI 1.1-6.9; p=0.03) and serum albumin <3.5 g/dl (90% vs. 66%; OR 3.2, 95% CI 1.0-10; p=0.045), while CXCR4 mutations associated with lower odds (44% vs. 82%; OR 0.15, 95% CI 0.06-0.37; p<0.001). Multivariate logistic regression analyses showed higher odds of VGPR at 6 months for b2-microglobulin ≥3 mg/l (21% vs. 3%; OR 3.3, 95% CI 1.1-10; p=0.04) and lower odds for serum IgM level ≥4,000 mg/dl (9% vs. 23%; OR 0.3, 95% CI 0.1-0.8; p=0.02). The median follow-up was 30 months, and the median PFS has not yet been reached. The 5-year PFS rate was 60% (95% CI 48-69%). In the multivariate Cox regression analysis, worse outcomes were seen with CXCR4 mutations (5-year PFS: 45% vs. 71%; HR 2.8, 95% CI 1.4-5.8; p=0.004) and serum albumin <3.5 g/dl (5-year PFS: 36% vs. 68%; HR 2.7, 95% CI 1.3-5.5; p=0.007). A novel PFS risk score was designed using CXCR4 mutational status and serum albumin (Figure), which divided patients into 3 distinct groups: low risk (no risk factors: 43%; 5-year PFS 81%), intermediate risk (1 risk factor: 46%; 5-year PFS 51%) and high risk (2 risk factors: 11%; median PFS 25 months). The PFS difference between groups was statistically significant (p<0.001). The PFS risk score showed consistent results when evaluating previously treated and untreated patients, as well as patients on and off clinical trials. Conclusion: Serum albumin and CXCR4 mutations emerge as important factors predictive of PR at 6 months and also prognostic of PFS in WM patients treated with ibrutinib. A novel PFS stratification tool that separates patients into 3 risk groups was established and would need further validation. Figure Disclosures Castillo: Abbvie: Research Funding; Janssen: Consultancy, Research Funding; Pharmacyclics: Consultancy, Research Funding; Beigene: Consultancy, Research Funding; TG Therapeutics: Research Funding. Hunter:Janssen: Consultancy. Treon:Pharmacyclics: Research Funding; BMS: Research Funding; Janssen: Consultancy.

Download Full-text

Predicting COVID-19 malignant progression with AI techniques

10.1101/2020.03.20.20037325 ◽

2020 ◽

Cited By ~ 15

Author(s):

Xiang Bai ◽

Cong Fang ◽

Yu Zhou ◽

Song Bai ◽

Zaiyi Liu ◽

...

Keyword(s):

Logistic Regression ◽

Deep Learning ◽

Treatment Strategy ◽

Protective Factor ◽

Prediction Models ◽

Laboratory Data ◽

Malignant Progression ◽

Medical Systems ◽

Multivariate Logistic Regression ◽

Critical Stage

AbstractBackground and purposeThe worldwide pandemic of coronavirus disease 2019 (COVID-19) greatly challenges public medical systems. With limited medical resources, the treatment priority is determined by the severity of patients. However, many mild outpatients quickly deteriorate into severe/critical stage. It is crucial to early identify them and give timely treatment for optimizing treatment strategy and reducing mortality. This study aims to establish an AI model to predict mild patients with potential malignant progression.MethodsA total of 133 consecutively mild COVID-19 patients at admission who was hospitalized in Wuhan Pulmonary Hospital from January 3 to February 13, 2020, were selected in this retrospective IRB-approved study. All mild patients were categorized into groups with or without malignant progression. The clinical and laboratory data at admission, the first CT, and the follow-up CT at the severe/critical stage of the two groups were compared. Both multivariate logistic regression and deep learning-based methods were used to build the prediction models, with their area under ROC curves (AUC) compared.ResultsMultivariate logistic regression depicted 6 risk factors for malignant progression: age >55years (OR 5.334, 95%CI 1.8-15.803), comorbid with hypertension (OR 5.093, 95%CI 1.236-20.986), a decrease of albumin (OR 4.01, 95%CI 1.216-13.223), a decrease of lymphocyte (OR 3.459, 95%CI 1.067-11.209), the progressive consolidation from CT1 to CTsevere (OR 1.235, 95%CI 1.018-1.498), and elevated HCRP (OR 1.015, 95%CI 1.002-1.029); and one protective factor: the presence of fibrosis at CT1 (OR 0.656, 95%CI 0.473-0.91). By combining the clinical data and the temporal information of the CT data, our deep learning-based models achieved the best AUC of 0.954, which outperformed logistic regression (AUC: 0.893),ConclusionsOur deep learning-based methods can identify the mild patients who are easy to deteriorate into severe/critical cases efficiently and accurately, which undoubtedly helps to optimize the treatment strategy, reduce mortality, and relieve the medical pressure.

Download Full-text