Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Use of feature importance statistics to accurately predict asthma attacks using machine learning: A cross-sectional cohort study of the US population

  • Alexander A. Huang,

    Roles Conceptualization, Methodology, Supervision, Writing – original draft, Writing – review & editing

    Affiliation Northwestern University Feinberg School of Medicine, Chicago, IL, United States of America

  • Samuel Y. Huang

    Roles Data curation, Formal analysis, Investigation, Methodology, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

    Huangs8@vcu.edu

    Affiliation Virginia Commonwealth University School of Medicine, Richmond, VA, United States of America

Abstract

Background

Asthma attacks are a major cause of morbidity and mortality in vulnerable populations, and identification of associations with asthma attacks is necessary to improve public awareness and the timely delivery of medical interventions.

Objective

The study aimed to identify feature importance of factors associated with asthma in a representative population of US adults.

Methods

A cross-sectional analysis was conducted using a modern, nationally representative cohort, the National Health and Nutrition Examination Surveys (NHANES 2017–2020). All adult patients greater than 18 years of age (total of 7,922 individuals) with information on asthma attacks were included in the study. Univariable regression was used to identify significant nutritional covariates to be included in a machine learning model and feature importance was reported. The acquisition and analysis of the data were authorized by the National Center for Health Statistics Ethics Review Board.

Results

7,922 patients met the inclusion criteria in this study. The machine learning model had 55 out of a total of 680 features that were found to be significant on univariate analysis (P<0.0001 used). In the XGBoost model the model had an Area Under the Receiver Operator Characteristic Curve (AUROC) = 0.737, Sensitivity = 0.960, NPV = 0.967. The top five highest ranked features by gain, a measure of the percentage contribution of the covariate to the overall model prediction, were Octanoic Acid intake as a Saturated Fatty Acid (SFA) (gm) (Gain = 8.8%), Eosinophil percent (Gain = 7.9%), BMXHIP–Hip Circumference (cm) (Gain = 7.2%), BMXHT–standing height (cm) (Gain = 6.2%) and HS C-Reactive Protein (mg/L) (Gain 6.1%).

Conclusion

Machine Learning models can additionally offer feature importance and additional statistics to help identify associations with asthma attacks.

Introduction

Asthma is a prevalent chronic respiratory disease that heterogeneously affects approximately 300 million people worldwide, causing recurrent episodes of coughing, wheezing, shortness of breath, and chest tightness [1]. Asthma attacks can be triggered by various environmental and genetic factors, including allergens, air pollution, stress, and infections [2]. These attacks can be debilitating, leading to hospitalization and even death in severe cases [3]. Moreover, asthma can significantly affect an individual’s quality of life, limiting their physical activities, causing sleep disturbance, and impairing their emotional well-being [4].

Asthma self-management plans aim to prevent attacks by helping patients identify when their asthma control is deteriorating and modifying their treatment accordingly [5]. However, predicting the risk of an asthma attack is currently a largely qualitative process, and there is no widely used algorithm to help identify patients at risk [6]. Most studies on the subject have used univariate regression modeling to identify risk factors for asthma attacks, which do not determine the predictive performance of an optimal combination of factors in individual patients [715].

To address these limitations, machine learning can be employed to identify factors that can cause asthma attacks. In this study, we used the National Health and Nutrition Examination Surveys (NHANES), a robust sample, to assess asthma attacks using an enhanced questionnaire. By employing XGBoost, which has strong backing from medical literature supporting it as a predictive machine learning model and a transparent machine learning process by the name of Shapely Additive Explanations (SHAP), our findings will identify hidden patterns and relationships that may not be evident through traditional statistical methods and provide valuable insights into the risk factors for asthma attacks and inform the development of personalized treatment plans for asthma patients [1618]. Shapley Additive Explanations (SHAP) is a feature importance method that provides insights into the contribution of individual features in predicting a target variable. It offers a unified framework for understanding the importance of features in a machine learning model, regardless of their correlation or dependence. When it comes to handling highly correlated variables, SHAP considers the collective contribution of features rather than attributing importance solely to individual variables. SHAP assigns each feature a Shapley value, which represents the average marginal contribution of a feature across all possible feature combinations. This approach helps address the issue of redundant or highly correlated variables. In the case of highly correlated variables, SHAP considers their joint contribution rather than duplicating importance [19]. The Shapley values are designed to allocate credit appropriately among the correlated features based on their actual impact on the model’s predictions. If two variables are highly correlated but provide similar information, their individual Shapley values may be reduced to reflect their shared contribution, rather than ranking both variables as highly important independently [20]. Through this approach, the study aimed to contribute to the broader understanding of asthma and facilitate improved management strategies for this condition.

Methods

The complex, multi-stage, cross-sectional cohort study utilized data from the publicly available National Health and Nutrition Examination Survey (NHANES) and included participants who completed questionnaire covering demographic information, dietary habits, exercise routines, and mental health, laboratory tests and physical exams. Participants were recruited for the NHANES program through a complicated, multistage probability sampling design. A stratified, multistage area probability design was used to select a representative sample from the civilian, non-institutionalized U.S. population for the sampling plan. Oversampling certain subgroups (such as low-income individuals, racial and ethnic minorities, and older adults) to obtain more precise estimates for these groups was done as part of the NHANES sampling design to ensure that the sample was representative of the U.S. population. Household interviews and in-person physical examinations in mobile examination centers were used for recruitment. The National Center for Health Statistics (NCHS) Ethics Review Board provided permission for the study’s data collection and analysis. Before analysis, all data, including medical records, survey responses, and demographic information, were de-identified to ensure participant anonymity. Authors did not have access to information that could identify participants during or after data collection. Furthermore, all participants provided verbal consent for their data to be made public.

Dataset and cohort selection

The NHANES 2017–2020 was developed by the NCHS, and it has been used to assess the health and nutritional status of the American population. To gather information regarding health, nutrition, and physical activity for the NHANES dataset, the Centers for Disease Control and Prevention (CDC) conducted a series of intricate, cross-sectional, multi-stage surveys on a nationally representative cohort of the population of the United States. Patients in the NHANES dataset who were adults (those over the age of 18) and who had completed the demographic, dietary, exercise, and mental health questionnaires as well as data from their physical and laboratory examinations were the subjects of our investigation.

Assessment of asthma attacks

The enhanced questionnaire used in NHANES asks patients about their experience with asthma over the past 12 months. Specifically, patients were asked whether they had experienced an episode of asthma or an asthma attack during this time period. Patients were asked “During the past 12 months, (have you/has SP) had an episode of asthma or an asthma attack?

Independent variable

In NHANES, the demographics, dietary, physical examination, laboratory, and medical questionnaire datasets contained potential model covariates. The NHANES dataset yielded a total of 680 covariates. An indicator of having experienced an asthma attack within the previous year was combined with all covariates.

Model construction and statistical analysis

Univariate logistic models used self-reported asthma attacks as the outcome to find covariates associated with asthma attacks. On univariate analysis, the final machine learning model had covariates with a p-value less than 0.0001. An initial filter on the dataset made use of univariable logistic models to make sure that each of the 680 covariates used in the machine learning models were strong independent covariates. Due to this initial filtering, physicians were also able to determine whether risk factors were clinically relevant. After performing initial filtering, relevant risk factors were identified by utilizing model importance statistics derived from machine-learning models.

The machine learning model XGBoost was used because of its widespread use in the literature and its improved predictive accuracy for healthcare predictions. Other studies using the NHANES cohort found that XGBoost was the most effective, offering the best balance between training efficiency, model accuracy, and transparency. A train:test set was used to calculate the final set of model fit parameters (80:20). In this study, the model fit parameters used were the area under the receiver operator characteristic curve (AUROC), sensitivity, specificity, positive predictive value, negative predictive value, prevalence, detection rate, detection prevalence, and balanced accuracy.

Model feature importance statistics and SHAP visualization

Model covariates were ranked according to Gain, Cover, and Frequency to identify factors that were associated with having an asthma attack within the previous year. The Gain is the feature’s relative contribution to the model. The total number of observations made regarding this feature is referred to as the Cover. The Frequency is the percentage of times a feature appears in the trees of the machine-learning model. The Gain statistic was chosen as the method for ranking features according to importance because it is easy to understand: the proportion of the final prediction that the covariate influenced. The continuous covariates with the strongest relationship between the potential risk factors and having an asthmatic attack within the past year were visualized using SHAP explanations.

Results

Table 1 shows the 7,922 patients that met the inclusion criteria in this study. Individuals reported mean Octanoic Acid intake as a Saturated Fatty Acid (SFA) of 0.42 grams a day in the general population (SD = 0.49), 0.54 grams a day among those that had an asthmatic attack (SD = 1.05), and 0.41 grams a day among those that did not have an asthmatic attack (SD = 0.44). Individuals had eosinophil percents of 2.78% (SD = 2.08) among the general population, 3.36% (SD = 2.73) among those that had an asthmatic attack, and 2.76% (SD = 2.05) among those that did not have an asthmatic attack. Individuals had an average hip circumference of 107.57 cm (SD = 14.89) among the general population, 115.16 cm (SD = 19.99) among those that had an asthmatic attack, and 107.24 cm (SD = 14.54) among those that did not have an asthmatic attack. Individuals had a standing height on average of 167.11 cm (SD = 9.95) among the general population, 164.73 cm (SD = 9.60) among those that had an asthmatic attack, and 167.22 cm (SD = 9.95) among those that did not have an asthmatic attack. Individuals had on average a HS C-Reactive protein level of 4.09 mg/L (SD = 7.85) among the general population, 6.08 mg/L (SD = 7.90) among those that had an asthmatic attack, and 4.00 mg/L (SD = 7.83) among those that did not have an asthmatic attack.

The machine learning model had 55 out of a total 680 features that were found to be significant on univariate analysis (P<0.0001 used). These were fitted into the XGBoost model and Fig 1 shows an AUROC = 0.737. Table 2 shows Sensitivity = 0.960, Specificity = 0.225 were observed. Table 3 shows that the top five highest ranked features by gain, a measure of the percentage contribution of the covariate to the overall model prediction, were Octanoic Acid intake as a SFA (gm) (Gain = 8.8%), Eosinophil percent (Gain = 7.9%), BMXHIP–Hip Circumference (cm) (Gain = 7.2%), BMXHT–standing height (cm) (Gain = 6.2%) and HS C-Reactive Protein (mg/L) (Gain 6.1%).

thumbnail
Fig 1. Receiver operator characteristic curve and model statistics.

The Receiver operating characteristic curve for the machine-learning model predicting whether the patient had an asthma attack within the past year or not. AUROC = 0.737 (P<0.001).

https://doi.org/10.1371/journal.pone.0288903.g001

In Fig 2, overall SHAP explanations can be seen for all the statistically significant covariates on univariable regression.

thumbnail
Fig 2. Overall SHAP explanations.

SHAP explanations, purple color representing higher values of the covariate while yellow representing lower values of the covariate. X-axis is the change in log-odds for reporting an asthma attack within the past year.

https://doi.org/10.1371/journal.pone.0288903.g002

In Fig 3 SHAP visualizations were conducted for the top four continuous covariates by overall SHAP explanations. We observed that if the patient told there doctor they had a sleeping disorder it increased their odds of having an asthmatic attack, increasing eosinophil percentage up to a asymptote of around 20% was associated with an increase risk of having an asthma attack the past year, reporting abdominal pain was associated with an increase risk of asthma attack, and increasing amounts of octanoic acid intake was associated with an increase risk of having an asthma attack within the past year.

thumbnail
Fig 3. SHAP explanations for the Top 4 continuous covariates sorted by overall SHAP explanations.

SHAP explanations, covariate value on the x-axis, change in log-odds on the y-axis, red line represents the relationship between the covariate and log-odds for Asthma attacks, each black dot represents an observation. Covariates: top left–Told doctor they had a sleeping disorder, top right–Eosinophil percent (%), bottom left–MCQ520 –Abdominal pain during past 12 month, bottom right–Octanoic Acid intake as a Saturated Fatty Acid (SFA) (gm).

https://doi.org/10.1371/journal.pone.0288903.g003

Discussion

In this cross sectional cohort of United States adults, a machine learning model to predict asthma attacks had great sensitivity and negative predictive value (AUROC = 0.737, sensitivity = 0.960, and NPV = 0.967) The greatest predictors for asthma attacks by SHAP explanation included if they told their doctor they had a sleeping disorder, eosinophil count, reporting abdominal pain in last 12 months, and octanoic intake in grams.

Our machine learning model contains relationships and findings that are in line with many other studies that have looked at factors associated with asthma attacks [2125]. Our model not only combines many of the associations found in literature, but demonstrates the importance and weights each item brings to the model and accounts for linear and curvilinear relationships. In our study the top 5 characteristics sorted by gain were concordant with current literature: there is strong epidemiological evidence that diet, eosinophils, hip circumference, standing height and C-reactive protein levels are associated with asthma [2231]. This alignment of these covariates with literature not only in statistical significance, but also in the direction of the relationships increase our confidence that the model accurately captures the physiological relationships of these factors. By using transparent machine learning tools, we can ensure that the model is detecting genuine signals within these covariates to predict asthma attacks, rather than simply replicating biases present in the dataset [3235]. The SHAP visualizations further support the increased predictive power of these non-parametric methods by demonstrating their ability to accurately capture the non-linear interactions between covariates, without overfitting the model to achieve greater accuracy [20, 3639].

By assisting patients in recognizing when they are increasing risk factors associated with asthma attacks, they can modify their treatment and routines with the aim to lower chances of attacks. However, there is currently no widely used guideline for identifying patients at risk, and predicting the likelihood of an asthma attack is largely a qualitative process. Many studies on the subject have used univariate regression modeling to identify asthma attack risk factors, but this method does not determine the optimal combination of factors for each patient’s predictive performance [4042].

However, a limited number of studies have attempted to combine various risk factors to develop a risk scoring algorithm and studies are increasingly utilizing machine learning in various capacities to study asthma. One benchmark study identified patients at risk of recurrent asthma attacks using the Optimum Patient Care Research Database, identifying various predictors for attacks and evaluated them using logistic regression. The study resulted in the development of an online asthma risk prediction tool for research and clinical purposes [4345]. However, this study only utilized logistic regression, which has been shown to be a poor classifier in cases of class imbalance.

Machine learning and artificial intelligence models have been effectively used to predict asthma attack [46, 47]. Prior studies have accurately predicted the presence of asthma attacks using machine-learning methods from a variety of datasets that consider physical parameters such as weight, blood pressure, and other metrics easily monitorable on wearable devices [48, 49]. Others look at prior exposure history as well as childhood development to predict potential of asthma exacerbation [5053]. These studies highlight the utility of machine learning models in identifying patients at risk for asthma attacks. In the past five years, numerous machine learning studies have been conducted for asthma management, but many of these studies have been limited by small sample sizes and lack of external validation, limiting their generalizability [47]. Our study, in contrast, utilized a large cohort, providing a more robust foundation for generalization and application of the developed algorithms. What our study adds to the literature is a large dataset (N = 7,922) that contains well representative data of US adults that looks at how a variety of nutrition and health examination data contribute to prediction of asthma attacks.

Many studies have also been pushing for increased transparency in modeling in general, but especially so in the field of pulmonology. Our study utilizes a novel paradigm recently described in literature that combines the commonly used XGBoost algorithm with SHAP to describe how the model is predicting each feature and how important it is [20].

The primary advantage of using this algorithmic method for covariate identification is the ability to systematically search through many potential factors, without relying on researcher judgment that may be influenced by personal biases [47]. This approach also enables the ranking of covariate importance using gain, cover and frequency statistics, allowing us to estimate the relative contribution of each covariate to asthma attack risk [20]. Additionally, SHAP visualizations can be employed after covariate selection and model building to ensure that each factor aligns with current literature on its association with asthma attacks, or to investigate any discrepancies and evaluate potential data-quality errors [18]. This approach strengthens the reliability and accuracy of our findings. The conclusion of the study highlights the potential of machine learning models in identifying strong and hidden associations related to predicting asthma attacks in patients within a one-year timeframe. By utilizing machine learning techniques, researchers can uncover patterns and relationships in the data that may not be immediately apparent through traditional statistical approaches. These models have the ability to analyze various factors and their interplay to provide insights into the occurrence of asthma attacks.

Limitations

Limitations of this machine-learning analysis include the cross-sectional nature and its ability to introduce bias. However, this was addressed by using training and testing sets to minimize overfitting. Additionally, SHAP visualizations provide a means for researchers to assess the plausibility of each covariate and determine whether the effects are due to true signal or noise that could contribute to a type-1 error. After considering the strengths and weaknesses of these methods, we propose that machine-learning can serve as a valuable initial step in identifying potential risk factors, which can then be further evaluated by clinicians based on the patient’s specific clinical presentation. NHANES, while being a robust and representative dataset of the US population, lacks detailed information on the severity of asthma attacks or validation measures. The dataset primarily relies on self-reported incidents of asthma attacks without specific measures of severity. This limitation restricts the studies ability to assess the severity of asthma attacks and validate the self-reported data. However, despite this limitation, the NHANES dataset provides a comprehensive examination of various factors associated with asthma attacks and offers valuable insights into the overall burden of asthma within the population. These findings, although not capturing the full spectrum of severity or validation, can still contribute important information to guide future research and inform public health interventions.

Conclusion

Machine learning models can be used to find strong and hidden associations for predicting if patient had asthma attack within the past year.

Supporting information

S1 Checklist. STROBE statement—checklist of items that should be included in reports of observational studies.

https://doi.org/10.1371/journal.pone.0288903.s001

(DOCX)

References

  1. 1. Anandan C, Nurmatov U, van Schayck OC, Sheikh A. Is the prevalence of asthma declining? Systematic review of epidemiological studies. Allergy. 2010;65(2):152–67. Epub 20091112. pmid:19912154.
  2. 2. Price D, Wilson AM, Chisholm A, Rigazio A, Burden A, Thomas M, et al. Predicting frequent asthma exacerbations using blood eosinophil count and other patient data routinely available in clinical practice. J Asthma Allergy. 2016;9:1–12. Epub 20160107. pmid:26793004; PubMed Central PMCID: PMC4708874.
  3. 3. Blakey JD, Price DB, Pizzichini E, Popov TA, Dimitrov BD, Postma DS, et al. Identifying Risk of Future Asthma Attacks Using UK Medical Record Data: A Respiratory Effectiveness Group Initiative. J Allergy Clin Immunol Pract. 2017;5(4):1015–24 e8. Epub 20161222. pmid:28017629.
  4. 4. Grana J, Preston S, McDermott PD, Hanchak NA. The use of administrative data to risk-stratify asthmatic patients. Am J Med Qual. 1997;12(2):113–9. pmid:9161058.
  5. 5. Mukherjee M, Stoddart A, Gupta RP, Nwaru BI, Farr A, Heaven M, et al. The epidemiology, healthcare and societal burden and costs of asthma in the UK and its member nations: analyses of standalone and linked national databases. BMC Med. 2016;14(1):113. Epub 20160829. pmid:27568881; PubMed Central PMCID: PMC5002970.
  6. 6. Loymans RJB, Debray TPA, Honkoop PJ, Termeer EH, Snoeck-Stroband JB, Schermer TRJ, et al. Exacerbations in Adults with Asthma: A Systematic Review and External Validation of Prediction Models. J Allergy Clin Immunol Pract. 2018;6(6):1942–52 e15. Epub 20180215. pmid:29454163.
  7. 7. Reddel HK, Taylor DR, Bateman ED, Boulet LP, Boushey HA, Busse WW, et al. An official American Thoracic Society/European Respiratory Society statement: asthma control and exacerbations: standardizing endpoints for clinical asthma trials and clinical practice. Am J Respir Crit Care Med. 2009;180(1):59–99. pmid:19535666.
  8. 8. Chen Y, Zhao A, Lyu J, Hu Y, Yin Y, Qu J, et al. Association of Parasomnia Symptoms with Risk of Childhood Asthma and the Role of Preterm Birth. Nat Sci Sleep. 2022;14:1559–73. Epub 20220903. pmid:36089999; PubMed Central PMCID: PMC9451036.
  9. 9. Duman Senol H, Topyildiz E, Ekici B, Gulen F, Demir E. Effectiveness and adverse reactions to subcutaneous immunotherapy in children with allergic rhinitis/asthma. Int J Pediatr Otorhinolaryngol. 2022;162:111292. Epub 20220819. pmid:36007303.
  10. 10. Faruqui Z, Thakker Z, Parveen D, Naik S, Urazbayeva M, Jain V, et al. Depression in Childhood Asthma vs. Adult-Onset Asthma: A Cross-Sectional Study from the National Health and Nutrition Examination Survey (NHANES). Children (Basel). 2022;9(12). Epub 20221123. pmid:36553241; PubMed Central PMCID: PMC9776752.
  11. 11. Gasiuniene E, Tamasauskiene L, Janulaityte I, Bjermer L, Sitkauskiene B. Clusters Based on Immune Markers in a Lithuanian Asthma Cohort Study. J Asthma. 2022:1–13. Epub 20221019. pmid:36260326.
  12. 12. Gill AS, Shipman P, Beswick DM, Howe H, Alt JA. Assessing Stakeholder Engagement for Outcomes-Based Research Among Patients With Chronic Rhinosinusitis and Asthma: A Survey-Based Investigation. Ann Otol Rhinol Laryngol. 2022:34894221129329. Epub 20221012. pmid:36226327.
  13. 13. Kraft M, Richardson M, Hallmark B, Billheimer D, Van den Berge M, Fabbri LM, et al. The role of small airway dysfunction in asthma control and exacerbations: a longitudinal, observational analysis using data from the ATLANTIS study. Lancet Respir Med. 2022;10(7):661–8. Epub 20220302. pmid:35247313.
  14. 14. Ogbu CE, Ravilla J, Okoli ML, Ahaiwe O, Ogbu SC, Kim ES, et al. Association of Depression, Poor Mental Health Status and Asthma Control Patterns in US Adults Using a Data-Reductive Latent Class Method. Cureus. 2023;15(1):e33966. Epub 20230119. pmid:36820113; PubMed Central PMCID: PMC9938719.
  15. 15. Ren J, Xu J, Zhang P, Bao Y. Prevalence and Risk Factors of Asthma in Preschool Children in Shanghai, China: A Cross-Sectional Study. Front Pediatr. 2021;9:793452. Epub 20220209. pmid:35223710; PubMed Central PMCID: PMC8864107.
  16. 16. Roque FS, Jensen PB, Schmock H, Dalgaard M, Andreatta M, Hansen T, et al. Using electronic patient records to discover disease correlations and stratify patient cohorts. PLoS Comput Biol. 2011;7(8):e1002141. Epub 20110825. pmid:21901084; PubMed Central PMCID: PMC3161904.
  17. 17. Naydenova E, Tsanas A, Howie S, Casals-Pascual C, De Vos M. The power of data mining in diagnosis of childhood pneumonia. J R Soc Interface. 2016;13(120). pmid:27466436; PubMed Central PMCID: PMC4971218.
  18. 18. Huang HH, Xu T, Yang J. Comparing logistic regression, support vector machines, and permanental classification methods in predicting hypertension. BMC Proc. 2014;8(Suppl 1):S96. Epub 20140617. pmid:25519351; PubMed Central PMCID: PMC4143639.
  19. 19. Lundberg S, Lee S-I. A Unified Approach to Interpreting Model Predictions 2017.
  20. 20. Huang AA, Huang SY. Increasing transparency in machine learning through bootstrap simulation and shapely additive explanations. PLoS One. 2023;18(2):e0281922. Epub 20230223. pmid:36821544; PubMed Central PMCID: PMC9949629.
  21. 21. Fletcher M, van der Molen T, Lenney W, Boucot I, Aggarwal B, Pizzichini E. Primary Care Management of Asthma Exacerbations or Attacks: Impact of the COVID-19 Pandemic. Adv Ther. 2022;39(4):1457–73. Epub 20220214. pmid:35157217; PubMed Central PMCID: PMC8853035.
  22. 22. Goddard BMM, Hutton A, Guilhermino M, McDonald VM. Parents’ Decision Making During Their Child’s Asthma Attack: Qualitative Systematic Review. J Asthma Allergy. 2022;15:1021–33. Epub 20220806. pmid:35967097; PubMed Central PMCID: PMC9365021.
  23. 23. Li X, Mao C, Pan Y. Effect of Routine Therapy Assisted by Physical Exercise on Pulmonary Function in Patients with Asthma in Stable Stage: A Systematic Review and Meta-analysis of Randomized Clinical Trials. Comput Math Methods Med. 2022;2022:2350297. Epub 20220614. pmid:35747133; PubMed Central PMCID: PMC9213151.
  24. 24. Liu X, Li J. Effects of acupoint herbal patching as an add-on to asthma medication during Sanfu days, as the hottest days in summer, on the acute attack, the immunological response, and the pulmonary function in asthmatic children: A meta-analysis. Medicine (Baltimore). 2022;101(37):e30247. pmid:36123839; PubMed Central PMCID: PMC9478312.
  25. 25. Sagmen SB, Eraslan BZ, Demirer E, Kiral N, Comert S. Small airway disease and asthma control. J Asthma. 2023:1–8. Epub 20230227. pmid:36847658.
  26. 26. Caliskaner Ozturk B, Borekci S, Yilmaz N, Gemicioglu B. Induced sputum eosinophilia and pulmonary function tests in ex-smokers with asthma controlled with an inhaled steroid and long-acting beta-agonist therapy. Tuberk Toraks. 2022;70(4):334–40. pmid:36537090.
  27. 27. de Marco R, Locatelli F, Sunyer J, Burney P. Differences in incidence of reported asthma related to age in men and women. A retrospective analysis of the data of the European Respiratory Health Survey. Am J Respir Crit Care Med. 2000;162(1):68–74. pmid:10903222.
  28. 28. Sundbom F, Lindberg E, Bjerg A, Forsberg B, Franklin K, Gunnbjornsdottir M, et al. Asthma symptoms and nasal congestion as independent risk factors for insomnia in a general population: results from the GA(2)LEN survey. Allergy. 2013;68(2):213–9. Epub 20121126. pmid:23176562.
  29. 29. To T, Vydykhan TN, Dell S, Tassoudji M, Harris JK. Is obesity associated with asthma in young children? J Pediatr. 2004;144(2):162–8. pmid:14760253.
  30. 30. Cepelis A, Brumpton BM, Laugsand LE, Dalen H, Langhammer A, Janszky I, et al. Asthma, asthma control and risk of acute myocardial infarction: HUNT study. Eur J Epidemiol. 2019;34(10):967–77. Epub 20190911. pmid:31512117.
  31. 31. Jin L, Gong H, Zhang Q. The clinical differences between cough variant asthma cells and humoral immunology indicators. Cell Mol Biol (Noisy-le-grand). 2022;68(4):188–93. Epub 20220430. pmid:35988279.
  32. 32. Bhandari M, Shahi TB, Siku B, Neupane A. Explanatory classification of CXR images into COVID-19, Pneumonia and Tuberculosis using deep learning and XAI. Comput Biol Med. 2022;150:106156. Epub 20221003. pmid:36228463; PubMed Central PMCID: PMC9549800.
  33. 33. El Bilali A, Abdeslam T, Ayoub N, Lamane H, Ezzaouini MA, Elbeltagi A. An interpretable machine learning approach based on DNN, SVR, Extra Tree, and XGBoost models for predicting daily pan evaporation. J Environ Manage. 2023;327:116890. Epub 20221129. pmid:36459782.
  34. 34. Fieggen J, Smith E, Arora L, Segal B. The role of machine learning in HIV risk prediction. Front Reprod Health. 2022;4:1062387. Epub 20221222. pmid:36619681; PubMed Central PMCID: PMC9815547.
  35. 35. Hu C, Li L, Li Y, Wang F, Hu B, Peng Z. Explainable Machine-Learning Model for Prediction of In-Hospital Mortality in Septic Patients Requiring Intensive Care Unit Readmission. Infect Dis Ther. 2022;11(4):1695–713. Epub 20220714. pmid:35835943; PubMed Central PMCID: PMC9282631.
  36. 36. Islam MS, Awal MA, Laboni JN, Pinki FT, Karmokar S, Mumenin KM, et al. HGSORF: Henry Gas Solubility Optimization-based Random Forest for C-Section prediction and XAI-based cause analysis. Comput Biol Med. 2022;147:105671. Epub 20220530. pmid:35660327.
  37. 37. Keir G, Hu W, Filippi CG, Ellenbogen L, Woldenberg R. Using artificial intelligence in medical school admissions screening to decrease inter- and intra-observer variability. JAMIA Open. 2023;6(1):ooad011. Epub 20230217. pmid:36819893; PubMed Central PMCID: PMC9936956.
  38. 38. Lee HG, Seo Y, Kim JH, Han SB, Im JH, Jung CY, et al. Machine learning model for predicting ciprofloxacin resistance and presence of ESBL in patients with UTI in the ED. Sci Rep. 2023;13(1):3282. Epub 20230225. pmid:36841917; PubMed Central PMCID: PMC9968289.
  39. 39. Mohammadifar A, Gholami H, Golzari S. Assessment of the uncertainty and interpretability of deep learning models for mapping soil salinity using DeepQuantreg and game theory. Sci Rep. 2022;12(1):15167. Epub 20220907. pmid:36071137; PubMed Central PMCID: PMC9452570.
  40. 40. Wang C, Qi Y, Chen Z. Explainable Gated Recurrent Unit to explore the effect of co-exposure to multiple air pollutants and meteorological conditions on mental health outcomes. Environ Int. 2023;171:107689. Epub 20221209. pmid:36508748.
  41. 41. Yin H, Sun Y, You Y, Notholt J, Palm M, Wang W, et al. Using machine learning approach to reproduce the measured feature and understand the model-to-measurement discrepancy of atmospheric formaldehyde. Sci Total Environ. 2022;851(Pt 2):158271. Epub 20220824. pmid:36028030.
  42. 42. Zhou Y, Yao X, Han W, Wang Y, Li Z, Li Y. Distinguishing apathy and depression in older adults with mild cognitive impairment using text, audio, and video based on multiclass classification and shapely additive explanations. Int J Geriatr Psychiatry. 2022;37(11). pmid:36284449.
  43. 43. Horne EMF, McLean S, Alsallakh MA, Davies GA, Price DB, Sheikh A, et al. Defining clinical subtypes of adult asthma using electronic health records: Analysis of a large UK primary care database with external validation. Int J Med Inform. 2023;170:104942. Epub 20221207. pmid:36529028.
  44. 44. Pace WD, Brandt E, Carter VA, Chang KL, Edwards CL, Evans A, et al. COPD Population in US Primary Care: Data From the Optimum Patient Care DARTNet Research Database and the Advancing the Patient Experience in COPD Registry. Ann Fam Med. 2022;20(4):319–27. pmid:35879087; PubMed Central PMCID: PMC9328699.
  45. 45. Soremekun S, Heaney LG, Skinner D, Bulathsinhala L, Carter V, Chaudhry I, et al. Asthma exacerbations are associated with a decline in lung function: a longitudinal population-based study. Thorax. 2022. Epub 20220803. pmid:35922128.
  46. 46. Tsang KCH, Pinnock H, Wilson AM, Salvi D, Shah SA. Predicting asthma attacks using connected mobile devices and machine learning: the AAMOS-00 observational study protocol. BMJ Open. 2022;12(10):e064166. Epub 20221003. pmid:36192103; PubMed Central PMCID: PMC9535155.
  47. 47. Tsang KCH, Pinnock H, Wilson AM, Shah SA. Application of Machine Learning Algorithms for Asthma Management with mHealth: A Clinical Review. J Asthma Allergy. 2022;15:855–73. Epub 20220629. pmid:35791395; PubMed Central PMCID: PMC9250768.
  48. 48. Alharbi ET, Nadeem F, Cherif A. Predictive models for personalized asthma attacks based on patient’s biosignals and environmental factors: a systematic review. BMC Med Inform Decis Mak. 2021;21(1):345. Epub 20211209. pmid:34886852; PubMed Central PMCID: PMC8656014.
  49. 49. Hussain Z, Shah SA, Mukherjee M, Sheikh A. Predicting the risk of asthma attacks in children, adolescents and adults: protocol for a machine learning algorithm derived from a primary care-based retrospective cohort. BMJ Open. 2020;10(7):e036099. Epub 20200723. pmid:32709646; PubMed Central PMCID: PMC7380838.
  50. 50. Kerr WT, Janio EA, Braesch CT, Le JM, Hori JM, Patel AB, et al. Identifying psychogenic seizures through comorbidities and medication history. Epilepsia. 2017;58(11):1852–60. Epub 20170912. pmid:28895657; PubMed Central PMCID: PMC5669805.
  51. 51. Lejeune S, Deschildre A, Le Rouzic O, Engelmann I, Dessein R, Pichavant M, et al. Childhood asthma heterogeneity at the era of precision medicine: Modulating the immune response or the microbiota for the management of asthma attack. Biochem Pharmacol. 2020;179:114046. Epub 20200522. pmid:32446884; PubMed Central PMCID: PMC7242211.
  52. 52. Tibble H, Tsanas A, Horne E, Horne R, Mizani M, Simpson CR, et al. Predicting asthma attacks in primary care: protocol for developing a machine learning-based prediction model. BMJ Open. 2019;9(7):e028375. Epub 20190709. pmid:31292179; PubMed Central PMCID: PMC6624024.
  53. 53. Toti G, Vilalta R, Lindner P, Lefer B, Macias C, Price D. Analysis of correlation between pediatric asthma exacerbation and exposure to pollutant mixtures with association rule mining. Artif Intell Med. 2016;74:44–52. Epub 20161125. pmid:27964802.