Diagnostic Accuracy of Web-Based COVID-19 Symptom Checkers: Comparison Study.
Ontology highlight
ABSTRACT: Background
A large number of web-based COVID-19 symptom checkers and chatbots have been developed; however, anecdotal evidence suggests that their conclusions are highly variable. To our knowledge, no study has evaluated the accuracy of COVID-19 symptom checkers in a statistically rigorous manner.Objective
The aim of this study is to evaluate and compare the diagnostic accuracies of web-based COVID-19 symptom checkers.Methods
We identified 10 web-based COVID-19 symptom checkers, all of which were included in the study. We evaluated the COVID-19 symptom checkers by assessing 50 COVID-19 case reports alongside 410 non-COVID-19 control cases. A bootstrapping method was used to counter the unbalanced sample sizes and obtain confidence intervals (CIs). Results are reported as sensitivity, specificity, F1 score, and Matthews correlation coefficient (MCC).Results
The classification task between COVID-19-positive and COVID-19-negative for "high risk" cases among the 460 test cases yielded (sorted by F1 score): Symptoma (F1=0.92, MCC=0.85), Infermedica (F1=0.80, MCC=0.61), US Centers for Disease Control and Prevention (CDC) (F1=0.71, MCC=0.30), Babylon (F1=0.70, MCC=0.29), Cleveland Clinic (F1=0.40, MCC=0.07), Providence (F1=0.40, MCC=0.05), Apple (F1=0.29, MCC=-0.10), Docyet (F1=0.27, MCC=0.29), Ada (F1=0.24, MCC=0.27) and Your.MD (F1=0.24, MCC=0.27). For "high risk" and "medium risk" combined the performance was: Symptoma (F1=0.91, MCC=0.83) Infermedica (F1=0.80, MCC=0.61), Cleveland Clinic (F1=0.76, MCC=0.47), Providence (F1=0.75, MCC=0.45), Your.MD (F1=0.72, MCC=0.33), CDC (F1=0.71, MCC=0.30), Babylon (F1=0.70, MCC=0.29), Apple (F1=0.70, MCC=0.25), Ada (F1=0.42, MCC=0.03), and Docyet (F1=0.27, MCC=0.29).Conclusions
We found that the number of correctly assessed COVID-19 and control cases varies considerably between symptom checkers, with different symptom checkers showing different strengths with respect to sensitivity and specificity. A good balance between sensitivity and specificity was only achieved by two symptom checkers.
SUBMITTER: Munsch N
PROVIDER: S-EPMC7541039 | biostudies-literature | 2020 Oct
REPOSITORIES: biostudies-literature
Publications
Diagnostic Accuracy of Web-Based COVID-19 Symptom Checkers: Comparison Study.
Munsch Nicolas N Martin Alistair A Gruarin Stefanie S Nateqi Jama J Abdarahmane Isselmou I Weingartner-Ortner Rafael R Knapp Bernhard B
Journal of medical Internet research 20201006 10
<h4>Background</h4>A large number of web-based COVID-19 symptom checkers and chatbots have been developed; however, anecdotal evidence suggests that their conclusions are highly variable. To our knowledge, no study has evaluated the accuracy of COVID-19 symptom checkers in a statistically rigorous manner.<h4>Objective</h4>The aim of this study is to evaluate and compare the diagnostic accuracies of web-based COVID-19 symptom checkers.<h4>Methods</h4>We identified 10 web-based COVID-19 symptom ch ...[more]
Similar Datasets
Project description:BackgroundIn the context of a deepening global shortage of health workers and, in particular, the COVID-19 pandemic, there is growing international interest in, and use of, online symptom checkers (OSCs). However, the evidence surrounding the triage and diagnostic accuracy of these tools remains inconclusive.ObjectiveThis systematic review aimed to summarize the existing peer-reviewed literature evaluating the triage accuracy (directing users to appropriate services based on their presenting symptoms) and diagnostic accuracy of OSCs aimed at lay users for general health concerns.MethodsSearches were conducted in MEDLINE, Embase, CINAHL, Health Management Information Consortium (HMIC), and Web of Science, as well as the citations of the studies selected for full-text screening. We included peer-reviewed studies published in English between January 1, 2010, and February 16, 2022, with a controlled and quantitative assessment of either or both triage and diagnostic accuracy of OSCs directed at lay users. We excluded tools supporting health care professionals, as well as disease- or specialty-specific OSCs. Screening and data extraction were carried out independently by 2 reviewers for each study. We performed a descriptive narrative synthesis.ResultsA total of 21,296 studies were identified, of which 14 (0.07%) were included. The included studies used clinical vignettes, medical records, or direct input by patients. Of the 14 studies, 6 (43%) reported on triage and diagnostic accuracy, 7 (50%) focused on triage accuracy, and 1 (7%) focused on diagnostic accuracy. These outcomes were assessed based on the diagnostic and triage recommendations attached to the vignette in the case of vignette studies or on those provided by nurses or general practitioners, including through face-to-face and telephone consultations. Both diagnostic accuracy and triage accuracy varied greatly among OSCs. Overall diagnostic accuracy was deemed to be low and was almost always lower than that of the comparator. Similarly, most of the studies (9/13, 69 %) showed suboptimal triage accuracy overall, with a few exceptions (4/13, 31%). The main variables affecting the levels of diagnostic and triage accuracy were the severity and urgency of the condition, the use of artificial intelligence algorithms, and demographic questions. However, the impact of each variable differed across tools and studies, making it difficult to draw any solid conclusions. All included studies had at least one area with unclear risk of bias according to the revised Quality Assessment of Diagnostic Accuracy Studies-2 tool.ConclusionsAlthough OSCs have potential to provide accessible and accurate health advice and triage recommendations to users, more research is needed to validate their triage and diagnostic accuracy before widescale adoption in community and health care settings. Future studies should aim to use a common methodology and agreed standard for evaluation to facilitate objective benchmarking and validation.Trial registrationPROSPERO CRD42020215210; https://tinyurl.com/3949zw83.
| S-EPMC10276326 | biostudies-literature
Project description:IntroductionThe development of machine learning models for symptom-based health checkers is a rapidly evolving area with significant implications for healthcare. Accurate and efficient diagnostic tools can enhance patient outcomes and optimize healthcare resources. This study focuses on evaluating and optimizing machine learning models using a dataset of 10 diseases and 9,572 samples.MethodsThe dataset was divided into training and testing sets to facilitate model training and evaluation. The following models were selected and optimized: Decision Tree, Random Forest, Naive Bayes, Logistic Regression and K-Nearest Neighbors. Evaluation metrics included accuracy, F1 scores, and 10-fold cross-validation. ROC-AUC and precision-recall curves were also utilized to assess model performance, particularly in scenarios with imbalanced datasets. Clinical vignettes were employed to gauge the real-world applicability of the models.ResultsThe performance of the models was evaluated using accuracy, F1 scores, and 10-fold cross-validation. The use of ROC-AUC curves revealed that model performance improved with increasing complexity. Precision-recall curves were particularly useful in evaluating model sensitivity in imbalanced dataset scenarios. Clinical vignettes demonstrated the robustness of the models in providing accurate diagnoses.DiscussionThe study underscores the importance of comprehensive model evaluation techniques. The use of clinical vignette testing and analysis of ROC-AUC and precision-recall curves are crucial in ensuring the reliability and sensitivity of symptom-based health checkers. These techniques provide a more nuanced understanding of model performance and highlight areas for further improvement.ConclusionThis study highlights the significance of employing diverse evaluation metrics and methods to ensure the robustness and accuracy of machine learning models in symptom-based health checkers. The integration of clinical vignettes and the analysis of ROC-AUC and precision-recall curves are essential steps in developing reliable and sensitive diagnostic tools.
| S-EPMC11483353 | biostudies-literature
Project description:BackgroundMedical self-diagnostic tools (or symptom checkers) are becoming an integral part of digital health and our daily lives, whereby patients are increasingly using them to identify the underlying causes of their symptoms. As such, it is essential to rigorously investigate and comprehensively report the diagnostic performance of symptom checkers using standard clinical and scientific approaches.ObjectiveThis study aims to evaluate and report the accuracies of a few known and new symptom checkers using a standard and transparent methodology, which allows the scientific community to cross-validate and reproduce the reported results, a step much needed in health informatics.MethodsWe propose a 4-stage experimentation methodology that capitalizes on the standard clinical vignette approach to evaluate 6 symptom checkers. To this end, we developed and peer-reviewed 400 vignettes, each approved by at least 5 out of 7 independent and experienced primary care physicians. To establish a frame of reference and interpret the results of symptom checkers accordingly, we further compared the best-performing symptom checker against 3 primary care physicians with an average experience of 16.6 (SD 9.42) years. To measure accuracy, we used 7 standard metrics, including M1 as a measure of a symptom checker's or a physician's ability to return a vignette's main diagnosis at the top of their differential list, F1-score as a trade-off measure between recall and precision, and Normalized Discounted Cumulative Gain (NDCG) as a measure of a differential list's ranking quality, among others.ResultsThe diagnostic accuracies of the 6 tested symptom checkers vary significantly. For instance, the differences in the M1, F1-score, and NDCG results between the best-performing and worst-performing symptom checkers or ranges were 65.3%, 39.2%, and 74.2%, respectively. The same was observed among the participating human physicians, whereby the M1, F1-score, and NDCG ranges were 22.8%, 15.3%, and 21.3%, respectively. When compared against each other, physicians outperformed the best-performing symptom checker by an average of 1.2% using F1-score, whereas the best-performing symptom checker outperformed physicians by averages of 10.2% and 25.1% using M1 and NDCG, respectively.ConclusionsThe performance variation between symptom checkers is substantial, suggesting that symptom checkers cannot be treated as a single entity. On a different note, the best-performing symptom checker was an artificial intelligence (AI)-based one, shedding light on the promise of AI in improving the diagnostic capabilities of symptom checkers, especially as AI keeps advancing exponentially.
| S-EPMC11091811 | biostudies-literature
Project description:BackgroundWeb-based symptom checkers are promising tools that provide help to patients seeking guidance on health problems. Many health organizations have started using them to enhance triage. Patients use the symptom checker to report their symptoms online and submit the report to the health care center through the system. Health care professionals (registered nurse, practical nurse, general physician, physiotherapist, etc) receive patient inquiries with urgency rating, decide on actions to be taken, and communicate these to the patients. The success of the adoption, however, depends on whether the tools can efficiently support health care professionals' workflow and achieve their support.ObjectiveThis study explores the factors influencing health care professionals' support for a web-based symptom checker for triage.MethodsData were collected through a web-based survey of 639 health care professionals using either of the two most used web-based symptom checkers in the Finnish public primary care. Linear regression models were fitted to study the associations between the study variables and health care professionals' support for the symptom checkers. In addition, the health care professionals' comments collected via survey were qualitatively analyzed to elicit additional insights about the benefits and challenges of the clinical use of symptom checkers.ResultsResults show that the perceived beneficial influence of the symptom checkers on health care professionals' work and the perceived usability of the tools were positively associated with professionals' support. The perceived benefits to patients and organizational support for use were positively associated, and threat to professionals' autonomy was negatively associated with health care professionals' support. These associations were, however, not independent of other factors included in the models. The influences on professionals' work were both positive and negative; the tools streamlined work by providing preliminary information on patients and reduced the number of phone calls, but they also created extra work as the professionals needed to call patients and ask clarifying questions. Managing time between the use of symptom checkers and other tasks was also challenging. Meanwhile, according to health care professionals' experience, the symptom checkers benefited patients as they received help quickly with a lower threshold for care.ConclusionsThe efficient use of symptom checkers for triage requires usable solutions that support health care professionals' work. High-quality information about the patients' conditions and an efficient way of communicating with patients are needed. Using a new eHealth tool also requires that health organizations and teams reorganize their workflows and work distributions to support clinical processes.
| S-EPMC9121216 | biostudies-literature
Project description:Symptom checkers are increasingly used to assess new symptoms and navigate the health care system. The aim of this study was to compare the accuracy of an artificial intelligence (AI)-based symptom checker (Ada) and physicians regarding the presence/absence of an inflammatory rheumatic disease (IRD). In this survey study, German-speaking physicians with prior rheumatology working experience were asked to determine IRD presence/absence and suggest diagnoses for 20 different real-world patient vignettes, which included only basic health and symptom-related medical history. IRD detection rate and suggested diagnoses of participants and Ada were compared to the gold standard, the final rheumatologists' diagnosis, reported on the discharge summary report. A total of 132 vignettes were completed by 33 physicians (mean rheumatology working experience 8.8 (SD 7.1) years). Ada's diagnostic accuracy (IRD) was significantly higher compared to physicians (70 vs 54%, p = 0.002) according to top diagnosis. Ada listed the correct diagnosis more often compared to physicians (54 vs 32%, p < 0.001) as top diagnosis as well as among the top 3 diagnoses (59 vs 42%, p < 0.001). Work experience was not related to suggesting the correct diagnosis or IRD status. Confined to basic health and symptom-related medical history, the diagnostic accuracy of physicians was lower compared to an AI-based symptom checker. These results highlight the potential of using symptom checkers early during the patient journey and importance of access to complete and sufficient patient information to establish a correct diagnosis.
| S-EPMC9548469 | biostudies-literature
Project description:BackgroundDiagnosis is a core component of effective health care, but misdiagnosis is common and can put patients at risk. Diagnostic decision support systems can play a role in improving diagnosis by physicians and other health care workers. Symptom checkers (SCs) have been designed to improve diagnosis and triage (ie, which level of care to seek) by patients.ObjectiveThe aim of this study was to evaluate the performance of the new large language model ChatGPT (versions 3.5 and 4.0), the widely used WebMD SC, and an SC developed by Ada Health in the diagnosis and triage of patients with urgent or emergent clinical problems compared with the final emergency department (ED) diagnoses and physician reviews.MethodsWe used previously collected, deidentified, self-report data from 40 patients presenting to an ED for care who used the Ada SC to record their symptoms prior to seeing the ED physician. Deidentified data were entered into ChatGPT versions 3.5 and 4.0 and WebMD by a research assistant blinded to diagnoses and triage. Diagnoses from all 4 systems were compared with the previously abstracted final diagnoses in the ED as well as with diagnoses and triage recommendations from three independent board-certified ED physicians who had blindly reviewed the self-report clinical data from Ada. Diagnostic accuracy was calculated as the proportion of the diagnoses from ChatGPT, Ada SC, WebMD SC, and the independent physicians that matched at least one ED diagnosis (stratified as top 1 or top 3). Triage accuracy was calculated as the number of recommendations from ChatGPT, WebMD, or Ada that agreed with at least 2 of the independent physicians or were rated "unsafe" or "too cautious."ResultsOverall, 30 and 37 cases had sufficient data for diagnostic and triage analysis, respectively. The rate of top-1 diagnosis matches for Ada, ChatGPT 3.5, ChatGPT 4.0, and WebMD was 9 (30%), 12 (40%), 10 (33%), and 12 (40%), respectively, with a mean rate of 47% for the physicians. The rate of top-3 diagnostic matches for Ada, ChatGPT 3.5, ChatGPT 4.0, and WebMD was 19 (63%), 19 (63%), 15 (50%), and 17 (57%), respectively, with a mean rate of 69% for physicians. The distribution of triage results for Ada was 62% (n=23) agree, 14% unsafe (n=5), and 24% (n=9) too cautious; that for ChatGPT 3.5 was 59% (n=22) agree, 41% (n=15) unsafe, and 0% (n=0) too cautious; that for ChatGPT 4.0 was 76% (n=28) agree, 22% (n=8) unsafe, and 3% (n=1) too cautious; and that for WebMD was 70% (n=26) agree, 19% (n=7) unsafe, and 11% (n=4) too cautious. The unsafe triage rate for ChatGPT 3.5 (41%) was significantly higher (P=.009) than that of Ada (14%).ConclusionsChatGPT 3.5 had high diagnostic accuracy but a high unsafe triage rate. ChatGPT 4.0 had the poorest diagnostic accuracy, but a lower unsafe triage rate and the highest triage agreement with the physicians. The Ada and WebMD SCs performed better overall than ChatGPT. Unsupervised patient use of ChatGPT for diagnosis and triage is not recommended without improvements to triage accuracy and extensive clinical evaluation.
| S-EPMC10582809 | biostudies-literature
Project description:BackgroundDegenerative cervical myelopathy (DCM) is a common debilitating condition resulting from degeneration of the cervical spine. While decompressive surgery can halt disease progression, existing spinal cord damage is often permanent, leaving patients with lifelong disability. Early surgery improves the likelihood of recovery, yet the average time from the onset of symptoms to correct diagnosis is over 2 years. The majority of delays occur initially, before and within primary care, mainly due to a lack of recognition. Symptom checkers are widely used by patients before medical consultation and can be useful for preliminary triage and diagnosis. Lack of recognition of DCM by symptom checkers may contribute to the delay in diagnosis.ObjectiveThe aims of this study were to investigate whether Web-based symptom checkers were able to recognize relevant symptoms of DCM, to characterize the DCM differential they returned , and to evaluate the diagnostic performance of recognized DCM symptoms.MethodsWe pooled classical DCM symptoms from leading review articles. These symptoms were entered into the algorithms used by the top 20 symptom checker websites (N=4; Google Search). The most widely cited symptom checker, WebMD, was used to characterize the differential diagnosis for DCM symptoms.ResultsA total of 31 classical DCM symptoms were identified, of which 45% (14/31) listed DCM as a differential and 10% (3/31) placed DCM in the top third of the differential. The mean differential rank for motor symptoms was significantly better than that for arthritic symptoms (P=.01) and the average differential rank for all symptoms (P=.048). The symptom checker WebMD performed best at recognizing DCM, placing the condition nearer to the top of the differential list (mean rank of 5.6) than either Healthline (rank of 12.9, P=.02) or Healthtools.AARP (rank of 15.5, P=.001). On WebMD, only one combination of symptoms resulted in DCM as the primary differential: neck, shoulder, and arm pain with hand weakness. Moreover, 151 differential diagnoses for DCM symptoms were recorded on WebMD. Multiple sclerosis and peripheral neuropathy were the most common differentials, shortlisted for 52% (16/31) and 32% (10/31) of the DCM symptoms, respectively.ConclusionsDCM symptoms are poorly identified by Web-based symptom checkers, which leads to a large differential of many other common conditions. While a diagnosis becomes more likely as the number of symptoms increases, this represents more advanced disease and will not support much-needed earlier diagnosis. Symptom checkers remain an attractive concept with potential. Further research is required to support their optimization.
| S-EPMC6330198 | biostudies-literature
Project description:ObjectivesThe aims of our study are firstly to investigate the diagnostic and triage performance of symptom checkers, secondly to assess their potential impact on healthcare utilisation and thirdly to investigate for variation in performance between systems.SettingPublicly available symptom checkers for patient use.ParticipantsPublicly available symptom-checkers were identified. A standardised set of 50 clinical vignettes were developed and systematically run through each system by a non-clinical researcher.Primary and secondary outcome measuresSystem accuracy was assessed by measuring the percentage of times the correct diagnosis was a) listed first, b) within the top five diagnoses listed and c) listed at all. The safety of the disposition advice was assessed by comparing it with national guidelines for each vignette.ResultsTwelve tools were identified and included. Mean diagnostic accuracy of the systems was poor, with the correct diagnosis being present in the top five diagnoses on 51.0% (Range 22.2 to 84.0%). Safety of disposition advice decreased with condition urgency (being 71.8% for emergency cases vs 87.3% for non-urgent cases). 51.0% of systems suggested additional resource utilisation above that recommended by national guidelines (range 18.0% to 61.2%). Both diagnostic accuracy and appropriate resource recommendation varied substantially between systems.ConclusionsThere is wide variation in performance between available symptom checkers and overall performance is significantly below what would be accepted in any other medical field, though some do achieve a good level of accuracy and safety of disposition. External validation and regulation are urgently required to ensure these public facing tools are safe.
| S-EPMC8282353 | biostudies-literature
Project description:BackgroundThe burden on healthcare systems is mounting continuously owing to population growth and aging, overuse of medical services, and the recent COVID-19 pandemic. This overload is also causing reduced healthcare quality and outcomes. One solution gaining momentum is the integration of intelligent self-assessment tools, known as symptom-checkers, into healthcare-providers' systems. To the best of our knowledge, no study so far has investigated the data-gathering capabilities of these tools, which represent a crucial resource for simulating doctors' skills in medical-interviews.ObjectivesThe goal of this study was to evaluate the data-gathering function of currently available chatbot symptom-checkers.MethodsWe evaluated 8 symptom-checkers using 28 clinical vignettes from the repository of MSD-Manual case studies. The mean number of predefined pertinent findings for each case was 31.8 ± 6.8. The vignettes were entered into the platforms by 3 medical students who simulated the role of the patient. For each conversation, we obtained the number of pertinent findings retrieved and the number of questions asked. We then calculated the recall-rates (pertinent-findings retrieved out of all predefined pertinent-findings), and efficiency-rates (pertinent-findings retrieved out of the number of questions asked) of data-gathering, and compared them between the platforms.ResultsThe overall recall rate for all symptom-checkers was 0.32(2,280/7,112;95 %CI 0.31-0.33) for all pertinent findings, 0.37(1,110/2,992;95 %CI 0.35-0.39) for present findings, and 0.28(1140/4120;95 %CI 0.26-0.29) for absent findings. Among the symptom-checkers, Kahun platform had the highest recall rate with 0.51(450/889;95 %CI 0.47-0.54). Out of 4,877 questions asked overall, 2,280 findings were gathered, yielding an efficiency rate of 0.46(95 %CI 0.45-0.48) across all platforms. Kahun was the most efficient tool 0.74 (95 %CI 0.70-0.77) without a statistically significant difference from Your.MD 0.69(95 %CI 0.65-0.73).ConclusionThe data-gathering performance of currently available symptom checkers is questionable. From among the tools available, Kahun demonstrated the best overall performance.
| S-EPMC9595333 | biostudies-literature
Project description:BackgroundThe diagnosis of inflammatory rheumatic diseases (IRDs) is often delayed due to unspecific symptoms and a shortage of rheumatologists. Digital diagnostic decision support systems (DDSSs) have the potential to expedite diagnosis and help patients navigate the health care system more efficiently.ObjectiveThe aim of this study was to assess the diagnostic accuracy of a mobile artificial intelligence (AI)-based symptom checker (Ada) and a web-based self-referral tool (Rheport) regarding IRDs.MethodsA prospective, multicenter, open-label, crossover randomized controlled trial was conducted with patients newly presenting to 3 rheumatology centers. Participants were randomly assigned to complete a symptom assessment using either Ada or Rheport. The primary outcome was the correct identification of IRDs by the DDSSs, defined as the presence of any IRD in the list of suggested diagnoses by Ada or achieving a prespecified threshold score with Rheport. The gold standard was the diagnosis made by rheumatologists.ResultsA total of 600 patients were included, among whom 214 (35.7%) were diagnosed with an IRD. Most frequent IRD was rheumatoid arthritis with 69 (11.5%) patients. Rheport's disease suggestion and Ada's top 1 (D1) and top 5 (D5) disease suggestions demonstrated overall diagnostic accuracies of 52%, 63%, and 58%, respectively, for IRDs. Rheport showed a sensitivity of 62% and a specificity of 47% for IRDs. Ada's D1 and D5 disease suggestions showed a sensitivity of 52% and 66%, respectively, and a specificity of 68% and 54%, respectively, concerning IRDs. Ada's diagnostic accuracy regarding individual diagnoses was heterogenous, and Ada performed considerably better in identifying rheumatoid arthritis in comparison to other diagnoses (D1: 42%; D5: 64%). The Cohen κ statistic of Rheport for agreement on any rheumatic disease diagnosis with Ada D1 was 0.15 (95% CI 0.08-0.18) and with Ada D5 was 0.08 (95% CI 0.00-0.16), indicating poor agreement for the presence of any rheumatic disease between the 2 DDSSs.ConclusionsTo our knowledge, this is the largest comparative DDSS trial with actual use of DDSSs by patients. The diagnostic accuracies of both DDSSs for IRDs were not promising in this high-prevalence patient population. DDSSs may lead to a misuse of scarce health care resources. Our results underscore the need for stringent regulation and drastic improvements to ensure the safety and efficacy of DDSSs.Trial registrationGerman Register of Clinical Trials DRKS00017642; https://drks.de/search/en/trial/DRKS00017642.
| S-EPMC11303907 | biostudies-literature