Objective To establish a machine learning-based risk prediction model of combined chronic obstructive pulmonary disease (COPD) with lung cancer, so as to explore the high risk factors for COPD patients with lung cancer and to lay the foundation for early detection of lung cancer risk in COPD patients. Methods A total of 154 patients from the Second Hospital of Dalian Medical University from 2010 to 2021 were retrospectively analyzed, including 99 patients in the COPD group and 55 patients in the COPD with lung cancer group. the chest high resolution computed tomography (HRCT) scans and pulmonary function test of each patient were acquired. The main analyses were as follow: (1) to valid the statistically differences of the basic information (such as age, body mass index, smoking index), laboratory test results, pulmonary function parameters and quantitative parameters of chest HRCT between the two groups; (2) to analyze the indicators of high risk factors for lung cancer in COPD patients using univariate and binary logistic regression (LR) methods; and (3) to establish the machine learning model (such as LR and Gaussian process) for COPD with lung cancer patients. Results Based on the statistical analysis and LR methods, decreased BMI, increased whole lung emphysema index, increased whole lung mean density, and increased percentage activity of exertional spirometry and prothrombin time were risk factors for COPD with lung cancer patients. Based on the machine learning prediction model for COPD with lung cancer patients, the area under the receiver operating characteristic curve for LR and Gaussian process were obtained as 0.88 using the soluble fragments of prothrombin time percentage activity, whole lung emphysema index, whole lung mean density, and forced vital capacity combined with neuron-specific enolase and cytokeratin 19 as features. Conclusion The prediction model of COPD with lung cancer patients using a machine learning approach can be used for early detection of lung cancer risk in COPD patients.
ObjectiveTo systematically review the efficacy of different non-pharmacological interventions for smoking cessation. MethodsPubMed, EMbase, The Cochrane Library, Web of Science, CBM, WanFang Data, VIP and CNKI databases were electronically searched to collect randomized controlled trials (RCTs) of different non-pharmacological interventions for smoking cessation from inception to November, 2021. Two reviewers independently screened literature, extracted data and assessed the risk of bias of included studies; then, network meta-analysis was performed by using Stata 15.1 software. ResultsA total of 27 RCTs involving 14 interventions were included. The results of the network meta-analysis showed that compared with conventional advice, video counseling (OR=2.34, 95%CI 1.32 to 4.15), mobile phone text message (OR=1.82, 95%CI 1.03 to 3.20), motivational interview (OR=2.00, 95%CI 1.11 to 3.59) and health education (OR=3.40, 95%CI 1.52 to 7.57) were higher in quitting rate (P<0.05). The sort results showed that health education was the most likely to be the best intervention (86.20%), followed by video consultation (74.10%). ConclusionCurrent evidence shows that the smoking cessation effects of health education, video counseling, telephone counseling, mobile phone text message and motivational interview. Among them, health education may be the best. Due to the limited quality and quantity of the included studies, more high-quality studies are needed to verify the above conclusion.
In response to the specific requirements of nutrition research, Schwingshackl’s research group developed the NutriGrade grading system, which independently assessed the quality of evidence in randomized controlled trials and cohort studies in nutrition, aiming to summarize the associations or effects between different nutritional factors and outcomes and meet the specific needs of evidence users. It has the advantages of novel classification, quantifiability, independence and pertinence, and it has better consistency, fairness, reliability and feasibility. Well-designed prospective cohort studies are more feasible in the field of nutrition than randomized controlled trials. The grading of the evidence quality for cohort studies included the following eight items: a) risk of bias, study quality, and study limitations; b) precision; c) heterogeneity; d) directness; e) publication bias; f) funding bias; g) effect size; and h) dose-response. Based on the evaluation results of the above items, the evidence quality could be divided into four grades: high (8-10), moderate (<8), low (<6), and very low (<4). The purpose of this paper was to introduce the basic principles, specific contents, and application methods of the NutriGrade grading system for cohort studies and cite examples to provide references for relevant researchers.