Disadvantages of Big Data in Healthcare Peer Review 2019
Review
Abstract
Groundwork: Although the potential of large data analytics for health care is well recognized, bear witness is lacking on its furnishings on public wellness.
Objective: The aim of this study was to assess the bear upon of the employ of big data analytics on people's health based on the wellness indicators and core priorities in the World Health Organization (WHO) General Programme of Piece of work 2019/2023 and the European Programme of Work (EPW), approved and adopted by its Member States, in addition to SARS-CoV-ii–related studies. Furthermore, we sought to place the most relevant challenges and opportunities of these tools with respect to people's health.
Methods: Vi databases (MEDLINE, Embase, Cochrane Database of Systematic Reviews via Cochrane Library, Web of Science, Scopus, and Epistemonikos) were searched from the inception date to September 21, 2020. Systematic reviews assessing the effects of big data analytics on health indicators were included. 2 authors independently performed screening, selection, data extraction, and quality assessment using the AMSTAR-two (A Measurement Tool to Assess Systematic Reviews 2) checklist.
Results: The literature search initially yielded 185 records, 35 of which met the inclusion criteria, involving more than v,000,000 patients. Most of the included studies used patient data collected from electronic wellness records, infirmary information systems, individual patient databases, and imaging datasets, and involved the use of large data analytics for noncommunicable diseases. "Probability of dying from whatsoever of cardiovascular, cancer, diabetes or chronic renal illness" and "suicide mortality rate" were the most commonly assessed health indicators and cadre priorities within the WHO General Programme of Work 2019/2023 and the EPW 2020/2025. Big data analytics have shown moderate to loftier accurateness for the diagnosis and prediction of complications of diabetes mellitus as well as for the diagnosis and classification of mental disorders; prediction of suicide attempts and behaviors; and the diagnosis, treatment, and prediction of important clinical outcomes of several chronic diseases. Conviction in the results was rated as "critically low" for 25 reviews, as "low" for seven reviews, and as "moderate" for 3 reviews. The well-nigh frequently identified challenges were institution of a well-designed and structured data source, and a secure, transparent, and standardized database for patient information.
Conclusions: Although the overall quality of included studies was limited, big data analytics has shown moderate to high accuracy for the diagnosis of certain diseases, improvement in managing chronic diseases, and support for prompt and real-time analyses of large sets of varied input data to diagnose and predict illness outcomes.
Trial Registration: International Prospective Register of Systematic Reviews (PROSPERO) CRD42020214048; https://world wide web.crd.york.ac.united kingdom/prospero/display_record.php?RecordID=214048
doi:10.2196/27275
Keywords
Introduction
Big data analytics tools handle complex datasets that traditional data processing systems cannot efficiently and economically store, manage, or process. Through the awarding of bogus intelligence (AI) algorithms and machine learning (ML), big data analytics has potential to revolutionize health intendance, supporting clinicians, providers, and policymakers for planning or implementing interventions [], faster affliction detection, therapeutic decision back up, outcome prediction, and increased personalized medicine, resulting in lower-cost, college-quality care with better outcomes [,].
In 2018, the World Wellness Arrangement (WHO) proposed the expedited 13th Full general Programme of Work (GPW13), which was approved and adopted past its 194 Member States, focusing on measurable impacts on people's health at the land level to transform public health with iii core features: enhanced universal wellness coverage, health emergencies protection, and amend health and well-being []. Forty-six outcome target indicators emerged from the GPW13, covering a range of health issues []. Big data analytics may help to support health policy controlling, accelerate the achievement of the GPW13 core priorities and targets, and guide the roadmap for the European region based on the European Programme of Piece of work (EPW) 2020/2025 [,].
Therefore, the aim of this report was to provide an overview of systematic reviews that assessed the furnishings of the use of big data analytics on people's health according to the WHO core features defined in the GPW13 and the EPW. Nosotros included complex reviews that assessed multiple interventions, dissimilar populations, and differing outcomes resulting from big information analytics on people's health, and identified the challenges, opportunities, and best practices for future research.
Methods
Study Pattern
This written report was designed to provide an overview of systematic reviews in accordance with guidelines from the Cochrane Handbook for Systematic Reviews of Interventions, along with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and the QUOROM (Quality of Reporting of Meta-analyses) guidelines [-]. The study protocol is published on PROSPERO (CRD42020214048).
Search Strategy
To identify records assessing the effect of big data analytics on people's health, aligned with the WHO health indicators defined in the GPW13 (), a comprehensive and systematic search was performed using six multidisciplinary databases from their inception to September 21, 2020. The search strategy was designed in collaboration with a senior librarian and is described in detail in .
References were imported into reference direction software (EndNote X9) and duplicates were removed. Unique records were uploaded onto the Covidence Platform (Veritas Health Innovation) for screening, data extraction, and quality assessment. A transmission search of reference lists was performed to supplement the search.
List of 46 Earth Health Organization health indicators defined at the Thirteenth General Programme of Work.
- Number of persons affected by disasters (per 100,000 population)
- Domestic full general authorities health expenditure (% of general government expenditure)
- Prevalence of stunting in children under 5 (%)
- Prevalence of wasting in children under v (%)
- Prevalence of overweight in children under 5 (%)
- Maternal mortality ratio (per 100,000 live births)
- Proportion of births attended by skilled wellness personnel (%)
- Under five mortality charge per unit (per chiliad live births)
- Neonatal mortality rate (per 1000 alive births)
- New HIV infections (per 1000 uninfected population)
- Tuberculosis incidence (per 100,000 population)
- Malaria incidence (per yard population at hazard)
- Hepatitis B incidence (measured past surface antigen [HBsAg] prevalence amongst children under v years)
- Number of people requiring interventions against neglected tropical diseases (NTDs)
- Probability of dying from any of cardiovascular affliction (CVD), cancer, diabetes, chronic renal affliction (CRD) (anile 30-70 years) (%)
- Suicide mortality rate (per 100,000 population)
- Coverage of treatment interventions for substance-use disorders (%)
- Total alcohol per capita consumption in adults aged >15 years (liters of pure alcohol)
- Road traffic bloodshed rate (per 100,000 population)
- Proportion of women (aged 15-49 years) having need for family planning satisfied with modern methods (%)
- Universal Wellness Coverage (UHC) Service Coverage Index
- Population with household expenditures on wellness >10% of total household expenditure or income (%)
- Mortality rate attributed to air pollution (per 100,000 population)
- Bloodshed rate attributed to exposure to dangerous water, sanitation, and hygiene (Wash) services (per 100,000 population)
- Mortality rate from unintentional poisoning (per 100,000 population)
- Prevalence of tobacco utilize in adults aged ≥15 years (%)
- Proportion of population covered by all vaccines included in national programs (diphtheria-tetanus-pertussis vaccine, measles-containing-vaccine second dose, pneumococcal conjugated vaccine) (%)
- Proportion of wellness facilities with essential medicines available and affordable on a sustainable basis (%)
- Density of health workers (doctors, nurse and midwives, pharmacists, dentists per x,000 population)
- International Health Regulations capacity and health emergency preparedness
- Proportion of bloodstream infections due to antimicrobial-resistant organisms (%)
- Proportion of children nether 5 years developmentally on track (wellness, learning, and psychosocial well-being) (%)
- Proportion of women (aged xv-49 years) subjected to violence by current or former intimate partner (%)
- Proportion of women (aged 15-49 years) who make their own decisions regarding sexual relations, contraceptive employ, and reproductive health care (%)
- Proportion of population using safely managed drinking-h2o services (%)
- Proportion of population using safely managed sanitation services and hand-washing facilities (%)
- Proportion of population with primary reliance on clean fuels (%)
- Annual mean concentrations of fine particulate matter (PM2.5) in urban areas (μg/m3)
- Proportion of children (anile ane-17 years) experiencing physical or psychological aggression (%)
- Vaccine coverage for epidemic-prone diseases
- Proportion of vulnerable people in fragile settings provided with essential health services (%)
- Prevalence of raised claret force per unit area in adults anile ≥18 years
- Effective policy/regulation for industrially produced trans-fat acids
- Prevalence of obesity (%)
- Number of cases of poliomyelitis caused by wild poliovirus
- Patterns of antibiotic consumption at the national level
Written report Choice
Peer-reviewed publications categorized as systematic reviews assessing the effects of big data analytics on whatsoever of the GPW13 and EPW health indicators and core priorities were included, regardless of linguistic communication and written report design. We only considered studies in which the search was performed in at least ii databases, and included a description of the search strategy and the methodology used for written report selection and data extraction. Nosotros only included studies that evaluated concrete relationships between the use of big data analytics and its outcome on people's lives, according to the WHO strategic priorities and indicators. Along with the 46 indicators listed in , we also included studies evaluating the use of big data during the COVID-19 pandemic. To identify gaps, we included reviews focusing on challenges, best practices, and short- and long-term opportunities related to large information technologies. Nonsystematic reviews, primary studies, opinions, short communications, nonscientific articles, conference abstracts, and reviews with big data inappropriately defined were excluded.
Although big data analysis is capable of handling large volumes of information, rather than focusing on the data volume/size, we focused on the process that defines big data analytics, which includes the following phases []: (1) data selection, (2) information preprocessing, (3) data transformation, (iv) AI/expert systems, and (5) understanding/assessment. The first three phases include subtasks such equally: (i) feature selection and extraction, (two) data cleaning, and (iii) data integration from multiple sources. The included studies covered all phases of the process. Title, abstract, and full-text screening were independently performed by ii reviewers using the inclusion criteria. Whatever disagreements were resolved past a third independent investigator.
Data Extraction
The post-obit data were extracted from the retrieved manufactures: publication information, journal proper noun and touch cistron, study characteristics, big information characteristics, outcomes, lessons and barriers for implementation, and principal limitations. Data were individually extracted by squad members and cross-checked for accuracy by a second investigator.
Assessment of Methodological Quality of Included Reviews
Two researchers independently assessed the studies using the AMSTAR 2 (A Measurement Tool to Assess Systematic Reviews ii) checklist, which includes the following critical domains, assessed in 16 items: protocol registered prior to review, capability of literature search, justification for excluded studies, adventure of bias in included studies, appropriateness of meta-analytic methods, consideration of bias risk when interpreting results, and assessing the presence and likely impact of publication bias []. Appropriateness to each appraisal feature was rated equally yes, no, fractional yes, not applicable, or unclear. Any conflict was resolved by a third party. Studies with a review protocol tracking number were analyzed. A concluding summary score was given to each included record, rated as "critically low," "depression," "moderate," or "high" [].
Information Synthesis
Results are reported in summary tables and through a narrative synthesis, grouping studies assessing the same disease or condition, and identifying challenges and opportunities. We also schematically correspond the evidence and gaps from these reviews as an overall synthesis.
Results
Reviews Retrieved
The search retrieved 1536 publications, 112 of which were duplicates. Most of the studies were excluded afterwards title and abstract analysis (n=1237), leaving 185 selected for total-text screening, and 35 [-] were ultimately included in the concluding analysis afterwards applying the eligibility criteria according to the QUOROM guidelines [] (). Reference list screening did not call up whatever additional review. One study under "awaiting classification" could not be retrieved.
Quality of Evidence in Individual Systematic Reviews
shows the detailed results of the quality assessment of the 35 systematic reviews. Overall, almost of the reviews (n=25) were rated with "critically depression" confidence in the results using the AMSTAR two criteria, with 7 rated "low" and 3 rated equally "moderate." None of the reviews accomplished a "high" rating. Common methodological drawbacks included omission of prospective protocol submission or publication, inappropriate search strategy, lack of independent and dual literature screening and data extraction, absence of caption for heterogeneity among the studies, unclear or no reasons for study exclusion, and lack of risk of bias cess.
No standard critical appraisement tools were mentioned. Among the 12 reviews that performed any quality assessment, the Quality Assessment of Diagnostic Accuracy Studies 2 tool was used in 4 reviews demonstrating an overall low risk of bias [,,,], whereas other tools assessed the risk of bias in studies non specifically aiming at diagnostic accurateness features. El Idrissi et al [] used their ain quality assessment tool and Luo et al [] used an adapted version of the Critical Appraisement Skills Program. Appraisal of the quality of evidence aligned with the Grading of Recommendations Assessment, Development and Evaluation method was reported in merely one review []. Many reviews did not evaluate bias.
Characteristics of Included Reviews
Summary features and chief findings of the 35 systematic reviews are presented in and , respectively. The included reviews were published in 34 dissimilar journals from 2007 to 2020. Well-nigh were published in English in a starting time-quartile periodical with an impact gene ranging from 0.977 to 17.679. They covered over 2501 principal studies, involving at to the lowest degree 5,000,000 individuals. Only three reviews included meta-analyses, and 1 included a randomized clinical trial; the others were based on cohort studies.
Data Sources and Purposes of Included Studies
Many reviews included data collected from electronic medical records, hospital data systems, or whatever databank that used individual patient information to create predictive models or evaluate commonage patterns [,,-,-,,-,,,,-]. Additionally, four reviews included primary studies based on imaging datasets and databanks, assessing unlike parameters of accuracy [,,,]. Other reviews focused on genetic databases [,], data from assisted reproductive technologies [], or publicly bachelor information [,,,]. Four studies lacked precision about the origin of the datasets used in their assay or did not specifically utilize patient information in the investigation [,,,].
The purposes of the reviews varied broadly. Generally, they (1) outlined AI applications in different medical specialties; (ii) analyzed features for the detection, prediction, or diagnosis of multiple diseases or conditions; or (iii) pinpointed challenges and opportunities.
WHO Indicators and Cadre Priorities
Most of the studies assessed the furnishings of large data analytics on noncommunicable diseases [-,,,,,,,,,, ,-]. Furthermore, three reviews covered mental wellness, associated with the indicator "suicide mortality rate" [,,]; 3 studies were related to the indicator "probability of dying from whatever of cardiovascular, cancer, diabetes, or chronic renal disease" [,,,,]; and two studies were related to the indicator "proportion of bloodstream infections due to antimicrobial-resistant organisms" [,]. One study described technology employ in disaster management and preparedness, covering the "number of persons afflicted by disasters" indicator [], and ane study was associated with the indicator "maternal mortality ratio" []. Overlap fabricated precise nomenclature into WHO health indicators challenging, and four studies could non be categorized because they mainly described challenges or opportunities in big information analytics [,] or because they were related to the COVID-nineteen pandemic [,].
Diseases or Weather Assessed
Diabetes Mellitus
AI tools associated with big data analytics in the care of patients with diabetes mellitus (DM) were assessed in six reviews that included 345 principal studies [,,,,]. Three studies reviewed AI in screening and diagnosing blazon i or blazon ii DM, providing varied ranges of accurateness, sensitivity, and specificity [,,]. Variables included systolic blood pressure, body mass index, triglyceride levels, and others. Two reviews covered DM control and the clinical management of DM patients [,]. One noted that techniques for diabetes self-management varied among the tools evaluated and reported mean values for its robust metrics []. The other evaluated the use of information-driven tools for predicting blood glucose dynamics and the affect of ML and information mining [], describing the input parameters used among data-driven analysis models. Even so, the authors of these reviews ended that achieving a methodologically precise predictive model is challenging and must consider multiple parameters.
Various studies assessed the power of large data analytics to predict individual DM complications such every bit hypoglycemia, nephropathy, and others [,,]. Supervised ML methods, conclusion copse, deep neural networks, random forests (RF) learning, and support vector auto (SVM) reportedly had the best outcomes for assessing complications. One review assessed deep learning–based algorithms in screening patients for diabetic retinopathy. Of eleven studies, viii reported sensitivity and specificity of 80.3% to 100% and 84%% to 99%, respectively; two reported accuracies of 78.7% and 81%; and one reported an area nether the receiver operating curve (AUC) of 0.955 [].
Mental Wellness
Five reviews reported on AI, data mining, and ML in psychiatry/psychology [,,,,], most commonly assessing these techniques in the diagnosis of mental disorders. Ii reviews assessed the utilise of ML algorithms for predicting suicidal behaviors. High levels of take a chance classification accuracy (typically higher than 90%) were reported in ii reviews, either for adult chief care patients or teenagers [,]. Although the review authors stated the potential of ML techniques in daily clinical practice, limitations were highlighted, including no external validation and reporting inconsistencies.
The utilise of ML algorithms for early on detection of psychiatric conditions was also reported [,]. ML was used to develop prediagnosis algorithms for constructing take a chance models to signal a patient's predisposition or chance for a psychiatric/psychological health effect, for predicting a diagnosis of newly identified patients, and to differentiate mental conditions with overlapping symptomatology. For studies using structural neuroimaging to classify bipolar diseases and other diagnoses, the accuracy ranged from 52.13% to 100%, whereas studies using serum biomarkers reported an accurateness ranging from 72.5% to 77.5%.
Merely one review used social media to generate analyzable data on the prevention, recognition, and support for severe mental illnesses []. The written report included broad descriptions of ML techniques and data types for detection, diagnosis, prognosis, treatment, support, and resulting public health implications. The authors highlighted the potential for monitoring well-being, and providing an ecologically and toll-efficient evaluation of customs mental health through social media and electronic records.
COVID-19
2 reviews reported the awarding of big information analytics and ML to better sympathise the electric current novel coronavirus pandemic [,]. One assessed data mining and ML techniques in diagnosing COVID-19 cases. Although the written report did non ascertain the all-time methodology to evaluate and detect potential cases, the authors noted an elevated frequency of conclusion tree models, naïve Bayes classifiers, and SVM algorithms used during previous pandemics.
Some other review focused on SARS-CoV-2 immunization, and proposed that AI could expedite vaccine discovery through studying the virus'south capabilities, virulence, and genome using genetic databanks. That report merged discussions of deep learning–based drug screening for predicting the interaction betwixt protein and ligands, and using imaging results linked to AI tools for detecting SARS-CoV-ii infections.
Oncology
Four studies described the utility of ML, computerized clinical determination systems, and deep learning in oncology [,,,]. Using computerized clinical determination support systems (DSS) significantly improves process outcomes in oncology []. A compelling example shows that initial decisions were modified in 31% of cases after consultation of clinical DSS, which consistently resulted in improved patient management. Furthermore, implementing clinical DSS led to an average cost reduction of Usa $17,000 for lung cancer patients. A remarkable workload decrease reportedly occurs when these systems are implemented in oncology facilities, leading to improved patient direction and adherence to guidelines [].
One written report evaluated ML techniques in a genomic written report of head and cervix cancers, and found a wide range of accuracy rates (56.7% to 99.4%) based on the employ of genomic information in prognostic prediction. Lastly, 2 studies reported accuracy levels ranging from 68% to 99.6% when using deep learning algorithms in the automatic detection of pulmonary nodules in computerized tomography images.
Cardiovascular and Neurological Weather
Half-dozen studies described the consequence of large data analytics in cardiology [,,,] and neurology [,]. One review assessed the employ of ML techniques for predicting cardiac arrest []. Different variables were used as predictors among individual studies, including electrocardiographic parameters, heart charge per unit variability, echocardiography, and others. Supervised ML techniques were most frequently practical to predict cardiac abort events, with articulate prove of regression techniques and SVM algorithms. The authors reported a hateful AUC of 0.76 for take a chance score development and efficiency evaluation [].
Similarly, two studies assessed the use of intelligent systems in diagnosing acute coronary syndrome and center failure [,], demonstrating loftier accuracy levels using several methods such every bit SVM, characteristic option, and neural networks. These studies likewise described useful clinical features for creating prediction and diagnostic models, such as patient clinical information, electrocardiogram characteristics, and cardiac biomarkers.
Scores to identify patients at higher risk to develop QT-interval prolongation have been developed, and predictive analytics incorporated into clinical conclusion support tools have been tested for their ability to alarm physicians of individuals who are at risk of or take QT-interval prolongation [].
Regarding stroke, ii systematic reviews evaluated using ML models for predicting outcomes and diagnosing cognitive ischemic events [,]. Generally, ML models were almost frequently associated with bloodshed prediction, functional outcomes, neurological deterioration, and quality of life. The diagnosis of ischemic stroke was associated with similar or better comparative accurateness for detecting big vessel occlusion compared with humans, depending on the AI algorithm employed []. RF algorithms had 68% sensitivity and over 80% specificity compared with humans. Analyses of convolutional neural network (CNN) algorithms were express, but systems using CNNs reported operation metrics on boilerplate 8% to ten% greater than those of ML employing RF, with upwardly to 85% hateful sensitivity for automated large vessel apoplexy detection. However, AI algorithm performance metrics used different standards, precluding objective comparison. Cadre and perfusion studies from RAPID-computed tomography and magnetic resonance imaging had the highest metrics for AI accuracy, above lxxx%, with some datasets showing 100% sensitivity to predict favorable perfusion mismatch. The authors noted several errors of AI utilise in diagnosing stroke [].
Miscellaneous Conditions
Several studies reported significant improvement in disease diagnosis and event prediction using big data analytics tools, including remarkable enhancement of sepsis prediction using ML techniques []. Another review provided moderate evidence that ML models tin can attain high operation standards in detecting wellness intendance–associated infections [].
One review focused on the diagnostic accuracy of AI systems in analyzing radiographic images for pulmonary tuberculosis, generally referring to development instead of clinical evaluation []. In studies assessing accuracy, the sensitivity ranged from 56% to 97%, specificity ranged from 36% to 95%, and the AUC ranged from 78% to 99%.
One review likewise assessed multiple sclerosis diagnosis. Amidst detection methodologies, dominion-based and natural linguistic communication processing methods were deemed to have superior diagnostic operation based of elevated accuracy and positive predictive value []. This study indicates that these methods take potential impacts for early recognition of the affliction, increasing quality of life, and assuasive prompt pharmacological and nonpharmacological intervention.
Asthma exacerbation events and predictive models for early on detection were evaluated in i review, which reached a pooled diagnostic power of 77% (95% CI 73%-80%) []. Amid the included studies, most models for predicting asthma development had less than 80% accuracy. None of the 42 studies modeled the reincidence of exacerbation events, and overall accurateness performance was considered inadequate. However, the authors encouraged creating models using large datasets to increase prediction accuracy levels. Logistic regression and Cox proportional hazard regression appeared to be the most commonly used methodologies. Gastric tissue affliction and the usability of deep learning techniques were evaluated in ane report []. CNN was the almost common model used for gastric problem nomenclature or detection. Additionally, residual neural network and fully convolutional network were considered to be advisable models for disease generation, classification, and sectionalisation.
Two reviews analyzed the utilise of big information analytics and AI in public health [,]. 1 listed the affect of continuous pharmacological exposure of meaning women, emphasizing that AI could improve popular understanding of drug effects on pregnancy, mainly through: (i) reliable clinical information disclosure, (2) adequate scientific enquiry pattern, and (iii) implementation of DSS []. Some other review assessing the use of large information in disaster preparedness evidenced that most existing methods are qualitative, covering the response phase of the disaster chain of events []. The utilized tools included data originating from geographic information systems, social media interfaces, and disaster prediction modeling studies.
Challenges and Opportunities
Two systematic reviews provided narrative evaluations of the challenges of large data analytics in health care [,]. Evidence from these two systematic reviews, and those from the other reviews, are summarized in .
Current challenges to use big data tools for peoples' health, and future perspectives and opportunities.
Current Challenges
1. Data structure: issues with fragmented information and incompatible or heterogeneous information formats
two. Information security: issues with privacy, lack of transparency, integrity, and inherent data construction
3. Data standardization: concerns with express interoperability, data obtention, mining, and sharing, along with language barriers
4. Inaccuracy: issues with inconsistencies, lack of precision, and data timeliness
5. Limited awareness of big data analytics capabilities amidst health managers and health intendance professionals
half-dozen. Lack of testify of large data analytics on the touch on on clinical outcomes for peoples' health
7. Lack of skills and grooming amidst professionals to collect, process, or extract data
8. Managerial bug: buying and government dilemma, along with information management, organizational, and financial issues
9. Regulatory, political, and legal concerns
10. Expenses with data storage and transfer
Future Perspectives and Opportunities
one. To improve the decision-making procedure with real-time analytics
2. To improve patient-centric health care and to enhance personalized medicine
3. To support early detection of diseases and prognostic assessment by predicting epidemics and pandemics, improving disease monitoring, implementing and tracking health behaviors, predicting patients' vulnerabilities
four. To improve data quality, structure, and accessibility by enabling the improvement of rapid conquering of big volumes and types of information, in a transparent style, and the improvement of information fault detection
v. To enable potential health intendance cost reduction
six. To improve quality of intendance past improving efficient health outcomes, reducing the waste material of resource, increasing productivity and performance, promoting chance reduction, and optimizing process management
seven. To provide improve forms to manage population health either through early detection of diseases or establishing ways to support health policy makers.
8. To enhance fraud detection
9. To enhance health-threat detection plans past governmental entities
x. To support the creation of new research hypotheses
Discussion
This overview is the get-go to assess the effects of large data analytics on the prioritized WHO indicators, which offers utility for noncommunicable diseases and the ongoing COVID-nineteen pandemic. Although the research question focused on the impact of big information analytics on people's health, studies assessing the impact on clinical outcomes are still deficient. Almost of the reviews assessed functioning values using big data tools and ML techniques, and demonstrated their applications in medical exercise. Most of the reviews were associated with the GPW13 indicator "probability of dying from any cardiovascular disease, cancer, diabetes, chronic respiratory disease." This indicator outranks others considering of the incidence, prevalence, premature mortality, and economic impact of these diseases []. Similarly, many reviews were related to "people requiring interventions confronting noncommunicable diseases." The included reviews in this study addressed many necessary health-related tasks; however, the quality of evidence was institute to be low to moderate, and studies assessing the affect on clinical outcomes are notably scarce.
The low to moderate quality of evidence suggests that large information analytics has moderate to loftier accurateness for the (1) diagnosis and prediction of complications of DM, (2) diagnosis of mental diseases, (3) prediction of suicidal behaviors, and (four) diagnosis of chronic diseases. Most studies presented performance values, although no study assessed whether big data analytics or ML could improve the early detection of specific diseases.
Clinical research and clinical trials significantly contribute to understanding the patterns and characteristics of diseases, as well as for improving detection of astute or chronic pathologies and to guide the development of novel medical interventions []. Still, experimental/theoretical investigations, mathematical approaches, and computer-based studies hinge on handling sample size limitations and performing information imputation [,]. Computer-driven analysis can easily handle missing data, examine variable mechanisms in complex systems, and apply essential tools for exploratory evaluations using voluminous input information. Big data analytics can execute an operation on/process data within microseconds afterwards generation of the dataset, assuasive for real-time follow up [,]. These studies and prospective applications could generate innovative knowledge and promote actionable insights; however, adapting, validating, and translating scientific data into applied medical protocols or evaluation studies is necessary.
Many systematic reviews reported elementary or inappropriate evaluation measures for the task at mitt. The near common metric used to evaluate the functioning of a nomenclature predictive model is accuracy, which is calculated as the proportion of correct predictions in the test prepare divided past the total number of predictions that were made on the test prepare. This metric is like shooting fish in a barrel to use and to translate, equally a unmarried number summarizes the model adequacy. However, accuracy values and error rate, which is only the complement of accuracy, are not acceptable for skewed or imbalanced classification tasks (ie, when the distribution of observations in the training dataset across the classes is not equal), because of the bias toward the majority form. When the distribution is slightly skewed, accurateness can still be a useful metric; however, when the distribution is severely skewed, accuracy becomes an unreliable mensurate of model performance.
For instance, in a binary classification task with a distribution of (95%, 5%) for the classes (eg, healthy vs ill), a "impaired classifier" that simply chooses the grade "good for you" for all instances volition accept 95% of accuracy in this chore, although the near important event in this task would be correctly classifying the "sick" class. Precision (also called the positive predictive value), which captures the fraction of correctly classified instances amid the instances predicted for a given form (eg, "sick"); recall or sensitivity, which captures the fraction of instances of a class (eg, "sick") that were correctly classified; and F-measure, the harmonic mean of precision and recall calculated per class of involvement, are more than robust metrics for several practical situations. The proper choice of an evaluation metric should be advisedly determined, as these indices ought to be used past regulatory bodies for screening tests and not for diagnostic reasoning []. The most of import issue is to choose the appropriate (most robust) performance metric given the particularities of each case.
Another pitfall identified among the included reviews was the lack of reporting the precise experimental protocols used for testing ML algorithms and the specific type of replication performed.
In that location is no formal tool for assessing quality and gamble of bias in big data studies. This is an area that is ripe for development. In , we summarize our recommendations for systematic reviews on the application of big data and ML for people's health based on our experience, the findings of this systematic review, and inspired by Cunha et al [].
High variability in the results was evident beyond different ML techniques and approaches among the 35 reviews, even for those assessing the same affliction or status. Indeed, designing large data analysis and ML experiments involves elevated model complexity and commonly requires testing of several modeling algorithms []. The diversity of big information tools and ML algorithms requires proper standardization of protocols and comparative approaches. Additionally, the process of tuning the hyperparameters of the algorithms is non uniformly reported. Important characteristics essential for replicability and external validation were non frequently available. Lastly, nigh of the studies provide little guidance to explain the results. Without knowing how and why the models achieve their results, applicability and trust of the models in real-world scenarios are severely compromised. Therefore, we urge the testing and assessment of supervised, unsupervised, and semisupervised methodologies, with explanation and estimation to justify the results. Moreover, nosotros encourage hyperparameter optimization to achieve adapted comeback of models, enhance model generalizations for untrained data, and avert overfitting to increase predictive accuracy.
Only two published systematic reviews evaluated the impact of big data analytics on the COVID-19 pandemic. Master studies on COVID-19 are lacking, which indicates an opportunity to apply big data and ML to this and future epidemics/pandemics [,]. As of November 30, 2020, many published protocols were retrieved through a standard search on PROSPERO. The titles of these review protocols showed an intention to evaluate ML tools in diagnosis and prediction, the affect of telemedicine using ML techniques, and the utilise of AI-based disease surveillance [].
Although DSS are an of import application of big data analytics and may do good patient care [-], but two reviews assessed such systems [,]. I focused on predictive analytics for identifying patients at risk of drug-induced QTc interval prolongation, discussing the efficacy of a DSS that has shown evidence of reduced prescriptions for QT interval–prolonging drugs. Similarly, one study exploring the impact of DSS on quality care in oncology showed that implementing these systems might positively touch physician-prescribing behaviors, health care costs, and clinician workload.
This overview of systematic reviews updates the available evidence from multiple primary studies intersecting reckoner science, engineering, medicine, and public wellness. We used a comprehensive search strategy (performed by an information specialist) with a predefined published protocol, precise inclusion criteria, rigorous data extraction, and quality cess of retrieved records. Nosotros avoided reporting bias through the dual and blinded examination of systematic reviews and past having one review author standardizing the extracted information.
Recommendations for systematic reviews on the application of large data and automobile learning for people's health.
- Cull an appropriate evaluation measure for the task and data characteristics, and justify your pick
Different evaluation measures such every bit accuracy, area nether the receiver operating feature bend, precision, call up, and F-measure capture different aspects of the task and are influenced past data characteristics such every bit skewness (ie, imbalance), sampling bias, etc. Choose your measures wisely and justify your choice based on the aforementioned aspects of the chore and the information.
- Ensure the employment of appropriate experimental protocols/design to guarantee generalization of the results
Authors should use experimental protocols based on cross-validation or multiple training/validation/test splits of the employed datasets with more than than one repetition of the experimental procedure. The objective of this criterion is to clarify whether the study assesses the capacity of generalization of each method compared in the experiments. The utilise of a single default split of the input dataset with only one preparation/test split does non fit this requirement. Repetitions are essential to demonstrate the generalization of the investigated methods for multiple training and examination sets, and to avoid any suspicion of a "lucky" (single) division that favors the authors' method.
- Properly melody, and explicitly study the tuning process and values of the hyperparameters of all compared methods
The effectiveness of big information solutions and motorcar-learning methods is highly affected past the option of the parameters of these methods (ie, parameter tuning). The wrong or improper choice of parameters may make a highly effective method exhibit very poor behavior in a given task. Ideally, the parameters should exist chosen for each specific task and dataset using a partition of the training set (ie, validation), which is different from the dataset used to railroad train and to test the model. This procedure is known as cross-validation on the training set or nested cross-validation.
Even if the tuning of all methods is properly executed, this should exist explicitly reported in the paper, with the exact values (or range of values) used for each parameter and the best choices used. When the tuning information is missing or absent-minded, it is impossible to make up one's mind whether the methods have been implemented appropriately and if they take achieved their maximum potential in a given job. Information technology is also impossible to assess whether the comparison is fair, as some methods may have been used at their maximum capacity and others non.
- Pay attention to the appropriate statistical tests
Authors should apply statistical significance tests to dissimilarity the compared strategies in their experimental evaluation. Statistical tests are essential to assess whether the functioning of the analyzed methods in the sample (ie, the considered datasets) is probable to reflect, with certain confidence, their bodily performance in the whole population. Equally such, they are key to support any claim of superiority of a detail method over others. Without such tests, the relative operation observed in the sample cannot, by whatever ways, be extrapolated to the population. The choice of the tests should also reflect the characteristics of the data (ie, determining whether the data follow a normal distribution).
- Brand the data and lawmaking freely available with proper documentation
Ane of the issues that hampers reproducibility of studies, and therefore scientific progress, is the lack of original implementation (with proper documentation) of the methods and techniques, and the unavailability of the original information used to examination the methods. Therefore, it is important to make all information, models, code, documentation, and other digital artifacts used in the inquiry available for others to reuse. The artifacts fabricated available must be sufficient to ensure that published results can exist accurately reproduced.
- Report other dimensions of the problem such every bit model costs (time) and potential for explainability
Effectiveness of the solutions, as captured by accuracy-oriented measures, is not the only dimension that should be evaluated. Indeed, if the effectiveness of the studied models is similar and sufficient for a given health-related application, other dimensions such as fourth dimension efficiency (or the costs) to train and deploy (exam) the models are essential to evaluate the practical applicability of such solutions. Some other dimension that may influence the decision for the practical apply of a big data or a machine-learning method in a existent applied situation is the ability to understand why the model has produced certain outputs (ie, explainability). Solutions such as those based on neural networks may be highly constructive when presented with huge amounts of information, but their training and deployment costs as well as their opaqueness may not make them the best choice for a given health-related awarding.
Notwithstanding, limitations be. The inferior quality scores based on the AMSTAR ii tool might reverberate incomplete reporting and lack of adherence to substandardized review methods. There is neither an established bias risk tool specifically for large information or ML studies nor any systematic way of presenting the findings of such studies. Furthermore, well-nigh studies provided a narrative clarification of results, requiring summarization. Nevertheless, all of the reviews were inspected by most authors, and the most relevant information were condensed in the text or in descriptive tables.
Big data analytics provide public health and wellness care with powerful instruments to gather and analyze large volumes of heterogeneous data. Although research in this field has been growing exponentially in the last decade, the overall quality of evidence is found to be low to moderate. High variability of results was observed beyond dissimilar ML techniques and approaches, even for the aforementioned affliction or condition. The diversity of big information tools and ML algorithms crave proper standardization of protocols and comparative approaches, and the process of tuning the hyperparameters of the algorithms is non uniformly reported. Important characteristics essential for replicability and external validation were non frequently available.
Additionally, the included reviews in this systematic review addressed different wellness-related tasks; however, studies assessing the impact on clinical outcomes remain scarce. Thus, prove of applicability in daily medical practice is nonetheless needed. Further studies should focus on how large data analytics affect clinical outcomes and on creating proper methodological guidelines for reporting big data/ML studies, too as using robust performance metrics to assess accuracy.
Acknowledgments
Nosotros highly appreciate the efforts provided by our experienced librarian Maria Björklund from Lund University, who kindly prepared the search strategy used in this inquiry. In add-on, nosotros thank Anneliese Arno (University Higher of London and Covidence Platform) for providing guidance in performing this research through Covidence. We also give thanks Raisa Eda de Resende, Edson Amaro Júnior, and Kaíque Amâncio Alvim for helping the group with information extraction and double-checking the input information.
Authors' Contributions
IJBdN, MM, MG, NAM, and DNO designed the study. HA, IW, and IJBdN performed first- and second-stage screening, and extracted the presented data. MM solved any disagreements. HA, IW, and IBdN carried out the quality assessment. IJBdN, MM, MG, and DNO drafted the manuscript and its final version. DNO and NAM are staff members of the WHO. The authors alone are responsible for the views expressed in this commodity and they do non necessarily stand for the decisions, policy, or views of the WHO.
Conflicts of Interest
None declared.
Multimedia Appendix 2
Quality assessment judgment using the AMSTAR ii tool.
DOCX File , 28 KB
Multimedia Appendix 4
Results and limitations of included systematic reviews.
DOCX File , 53 KB
Abbreviations
| AI: artificial intelligence |
| AUC: area under the receiver operating characteristic curve |
| AMSTAR 2: A Measurement Tool to Assess Systematic Reviews two |
| CNN: convolutional neural network |
| DM: diabetes mellitus |
| DSS: decision support system |
| EPW: European Programme of Piece of work |
| GPW13: Thirteenth General Programme of Work |
| ML: car learning |
| PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-analyses |
| QUOROM: Quality of Reporting of Meta-analyses |
| RF: random woods |
| SVM: back up vector motorcar |
| WHO: Globe Health Organization |
Edited by R Kukafka, Thou Eysenbach; submitted 19.01.21; peer-reviewed by Y Mejova, A Benis; comments to author 09.02.21; revised version received 19.02.21; accustomed 24.03.21; published 13.04.21
Copyright©Israel Júnior Borges do Nascimento, Milena Soriano Marcolino, Hebatullah Mohamed Abdulazeem, Ishanka Weerasekara, Natasha Azzopardi-Muscat, Marcos André Gonçalves, David Novillo-Ortiz. Originally published in the Periodical of Medical Internet Inquiry (http://www.jmir.org), 13.04.2021.
This is an open-access article distributed nether the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/four.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, starting time published in the Journal of Medical Cyberspace Inquiry, is properly cited. The consummate bibliographic information, a link to the original publication on http://www.jmir.org/, every bit well as this copyright and license information must exist included.
Source: https://www.jmir.org/2021/4/e27275/
0 Response to "Disadvantages of Big Data in Healthcare Peer Review 2019"
Post a Comment