Clinical Utility and Functionality of an Artificial Intelligence-Based App to Predict Mortality in COVID-19: Mixed Methods Analysis

Quality of interaction between clinicians and artificial intelligence systems. A systematic review

Introduction

Artificial intelligence (AI) has the potential to improve healthcare quality when thoughtfully integrated into clinical practice. Current evaluations of AI solutions tend to focus solely on model performance. There is a critical knowledge gap in the assessment of AI-clinician interactions. We systematically reviewed existing literature to identify interaction traits that can be used to assess the quality of AI-clinician interactions.

Methods

We performed a systematic review of published studies to June 2022 that reported elements of interactions that impacted the relationship between clinicians and AI-enabled clinical decision support systems. Due to study heterogeneity, we conducted a narrative synthesis of the different interaction traits identified from this review. Two study authors categorised the AI-clinician interaction traits based on their shared constructs independently. After the independent categorisation, both authors engaged in a discussion to finalise the categories.

Results

From 34 included studies, we identified 210 interaction traits. The most common interaction traits included usefulness, ease of use, trust, satisfaction, willingness to use and usability. After removing duplicate or redundant traits, 90 unique interaction traits were identified. Unique interaction traits were then classified into seven categories: usability and user experience, system performance, clinician trust and acceptance, impact on patient care, communication, ethical and professional concerns, and clinician engagement and workflow.

Discussion

We identified seven categories of interaction traits between clinicians and AI systems. The proposed categories may serve as a foundation for a framework assessing the quality of AI-clinician interactions.

Trust and acceptability of data-driven clinical recommendations in everyday practice: A scoping review

BACKGROUND

Increasing attention is being given to the analysis of large health datasets to derive new clinical decision support systems (CDSS). However, few data-driven CDSS are being adopted into clinical practice. Trust in these tools is believed to be fundamental for acceptance and uptake but to date little attention has been given to defining or evaluating trust in clinical settings.

OBJECTIVES

A scoping review was conducted to explore how and where acceptability and trustworthiness of data-driven CDSS have been assessed from the health professional's perspective.

METHODS

Medline, Embase, PsycInfo, Web of Science, Scopus, ACM Digital, IEEE Xplore and Google Scholar were searched in March 2022 using terms expanded from: "data-driven" AND "clinical decision support" AND "acceptability". Included studies focused on healthcare practitioner-facing data-driven CDSS, relating directly to clinical care. They included trust or a proxy as an outcome, or in the discussion. The preferred reporting items for systematic reviews and meta-analyses extension for scoping reviews (PRISMA-ScR) is followed in the reporting of this review.

RESULTS

3291 papers were screened, with 85 primary research studies eligible for inclusion. Studies covered a diverse range of clinical specialisms and intended contexts, but hypothetical systems (24) outnumbered those in clinical use (18). Twenty-five studies measured trust, via a wide variety of quantitative, qualitative and mixed methods. A further 24 discussed themes of trust without it being explicitly evaluated, and from these, themes of transparency, explainability, and supporting evidence were identified as factors influencing healthcare practitioner trust in data-driven CDSS.

CONCLUSION

There is a growing body of research on data-driven CDSS, but few studies have explored stakeholder perceptions in depth, with limited focused research on trustworthiness. Further research on healthcare practitioner acceptance, including requirements for transparency and explainability, should inform clinical implementation.

The use of artificial intelligence for delivery of essential health services across WHO regions: a scoping review

Background

Artificial intelligence (AI) is a broad outlet of computer science aimed at constructing machines capable of simulating and performing tasks usually done by human beings. The aim of this scoping review is to map existing evidence on the use of AI in the delivery of medical care.

Methods

We searched PubMed and Scopus in March 2022, screened identified records for eligibility, assessed full texts of potentially eligible publications, and extracted data from included studies in duplicate, resolving differences through discussion, arbitration, and consensus. We then conducted a narrative synthesis of extracted data.

Results

Several AI methods have been used to detect, diagnose, classify, manage, treat, and monitor the prognosis of various health issues. These AI models have been used in various health conditions, including communicable diseases, non-communicable diseases, and mental health.

Conclusions

Presently available evidence shows that AI models, predominantly deep learning, and machine learning, can significantly advance medical care delivery regarding the detection, diagnosis, management, and monitoring the prognosis of different illnesses.

Application of machine learning approach in emergency department to support clinical decision making for SARS-CoV-2 infected patients

To support physicians in clinical decision process on patients affected by Coronavirus Disease 2019 (COVID-19) in areas with a low vaccination rate, we devised and evaluated the performances of several machine learning (ML) classifiers fed with readily available clinical and laboratory data. Our observational retrospective study collected data from a cohort of 779 COVID-19 patients presenting to three hospitals of the Lazio-Abruzzo area (Italy). Based on a different selection of clinical and respiratory (ROX index and PaO2/FiO2 ratio) variables, we devised an AI-driven tool to predict safe discharge from ED, disease severity and mortality during hospitalization. To predict safe discharge our best classifier is an RF integrated with ROX index that reached AUC of 0.96. To predict disease severity the best classifier was an RF integrated with ROX index that reached an AUC of 0.91. For mortality prediction the best classifier was an RF integrated with ROX index, that reached an AUC of 0.91. The results obtained thanks to our algorithms are consistent with the scientific literature an accomplish significant performances to forecast safe discharge from ED and severe clinical course of COVID-19.

An Assessment of How Clinicians and Staff Members Use a Diabetes Artificial Intelligence Prediction Tool: Mixed Methods Study

BACKGROUND

Nearly one-third of patients with diabetes are poorly controlled (hemoglobin A1c≥9%). Identifying at-risk individuals and providing them with effective treatment is an important strategy for preventing poor control.

OBJECTIVE

This study aims to assess how clinicians and staff members would use a clinical decision support tool based on artificial intelligence (AI) and identify factors that affect adoption.

METHODS

This was a mixed methods study that combined semistructured interviews and surveys to assess the perceived usefulness and ease of use, intent to use, and factors affecting tool adoption. We recruited clinicians and staff members from practices that manage diabetes. During the interviews, participants reviewed a sample electronic health record alert and were informed that the tool uses AI to identify those at high risk for poor control. Participants discussed how they would use the tool, whether it would contribute to care, and the factors affecting its implementation. In a survey, participants reported their demographics; rank-ordered factors influencing the adoption of the tool; and reported their perception of the tool's usefulness as well as their intent to use, ease of use, and organizational support for use. Qualitative data were analyzed using a thematic content analysis approach. We used descriptive statistics to report demographics and analyze the findings of the survey.

RESULTS

In total, 22 individuals participated in the study. Two-thirds (14/22, 63%) of respondents were physicians. Overall, 36% (8/22) of respondents worked in academic health centers, whereas 27% (6/22) of respondents worked in federally qualified health centers. The interviews identified several themes: this tool has the potential to be useful because it provides information that is not currently available and can make care more efficient and effective; clinicians and staff members were concerned about how the tool affects patient-oriented outcomes and clinical workflows; adoption of the tool is dependent on its validation, transparency, actionability, and design and could be increased with changes to the interface and usability; and implementation would require buy-in and need to be tailored to the demands and resources of clinics and communities. Survey findings supported these themes, as 77% (17/22) of participants somewhat, moderately, or strongly agreed that they would use the tool, whereas these figures were 82% (18/22) for usefulness, 82% (18/22) for ease of use, and 68% (15/22) for clinic support. The 2 highest ranked factors affecting adoption were whether the tool improves health and the accuracy of the tool.

CONCLUSIONS

Most participants found the tool to be easy to use and useful, although they had concerns about alert fatigue, bias, and transparency. These data will be used to enhance the design of an AI tool.

Incubation period, clinical and lung CT features for early prediction of COVID-19 deterioration: development and internal verification of a risk model

Background

Most severe, critical, or mortal COVID-19 cases often had a relatively stable period before their status worsened. We developed a deterioration risk model of COVID-19 (DRM-COVID-19) to predict exacerbation risk and optimize disease management on admission.

Method

We conducted a multicenter retrospective cohort study with 239 confirmed symptomatic COVID-19 patients. A combination of the least absolute shrinkage and selection operator (LASSO), change-in-estimate (CIE) screened out independent risk factors for the multivariate logistic regression model (DRM-COVID-19) from 44 variables, including epidemiological, demographic, clinical, and lung CT features. The compound study endpoint was progression to severe, critical, or mortal status. Additionally, the model's performance was evaluated for discrimination, accuracy, calibration, and clinical utility, through internal validation using bootstrap resampling (1000 times). We used a nomogram and a network platform for model visualization.

Results

In the cohort study, 62 cases reached the compound endpoint, including 42 severe, 18 critical, and two mortal cases. DRM-COVID-19 included six factors: dyspnea [odds ratio (OR) 4.89;confidence interval (95% CI) 1.53–15.80], incubation period (OR 0.83; 95% CI 0.68–0.99), number of comorbidities (OR 1.76; 95% CI 1.03–3.05), D-dimer (OR 7.05; 95% CI, 1.35–45.7), C-reactive protein (OR 1.06; 95% CI 1.02–1.1), and semi-quantitative CT score (OR 1.50; 95% CI 1.27–1.82). The model showed good fitting (Hosmer–Lemeshow goodness, X²(8) = 7.0194, P = 0.53), high discrimination (the area under the receiver operating characteristic curve, AUROC, 0.971; 95% CI, 0.949–0.992), precision (Brier score = 0.051) as well as excellent calibration and clinical benefits. The precision-recall (PR) curve showed excellent classification performance of the model (AUC_PR = 0.934). We prepared a nomogram and a freely available online prediction platform (https://deterioration-risk-model-of-covid-19.shinyapps.io/DRMapp/).

Conclusion

We developed a predictive model, which includes the including incubation period along with clinical and lung CT features. The model presented satisfactory prediction and discrimination performance for COVID-19 patients who might progress from mild or moderate to severe or critical on admission, improving the clinical prognosis and optimizing the medical resources.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12890-022-01986-0.