The global aging is one of the most important health problems worldwide.1 Meanwhile, little evidence showed that increasing longevity was being accompanied by an extended period of health.2 People over 65 years old are expected to spend the rest of their life under a limiting physical and mental condition. A number of health problems may occur with aging, which includes, but not limited to, frailty, sarcopenia, and multiple comorbidities.3 Health problems are not only diseases themselves, but also indicate potential health risks. And they are closely related to an increased risk of mobility limitations, falls, loss of independence, disability, metabolic disorders, and mortality.4 In order to fostering healthy aging, WHO had proposed the World report on aging and health.5 Since then, more and more researches have been done to solve the issue.

Early diagnosis has already become common consensus in various kinds of diseases. For instance, early diagnosis of atrial fibrillation (AF) would reduce future hospitalizations, morbidity, mortality and AF-related strokes.6,7 Similar results were seen in other health problems, including diabetes,8,9 Parkinson Disease (PD)10,11 and COPD.12 However, due to costs, transportation, COVID-19-related social isolation and other issues, regular hospital visits might be difficult for some old adults.13,14 In these conditions, it could be difficult to make early diagnosis of the majority of health problems.

Electronic-devices (e-devices) might be potential in early detection and management of some health problems among old adults. These devices present a promising solution for remote monitoring, assessment and diagnosis through ambient assisted living.15 In fact, e-devices have been proved applicable in multiple health problems.16 Yet, not all the e-devices based applications are suitable for the elderly to use independently at home. For example, Christiansen et al. developed a new system based on smartphone where fingertip capillary blood samples were required in order to monitor blood glucose.17 Pérez-Ros and his colleagues made detection of falls possible, however, via a wearable device that needed additional assists during wearing.18 Briefly speaking, professional requirements and relatively complex operation are the main limitations, which made some of the e-devices based systems ‘household unavailable’.

Current e-devices offer novel, cost-effective and portable approaches for detection of health problems in older adults at home.19 And Kristoffersson et al. made a qualitative synthesis on wearable body sensors for health monitoring in their review.16 However, to our knowledge, there is no quantitative systematic review for the diagnosis validity on these novel approaches. Therefore, the aim of this systematic review and meta-analysis for diagnostic test accuracy studies is to find out the overall effectiveness and ability of the household available e-devices on diagnosing different health problems in old adults. In addition, this systematic review aimed at a range of diagnoses rather than disease-specific diagnostic accuracy.

This systematic review was being reported according to the Preferred Reporting Items for a Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) guidelines.20 At the same time, a search for previous and ongoing systematic reviews similar to this study was conducted in order to avoid duplication.

Seven online databases were searched for publications conducted until June 2022, namely: PubMed, Cochrane, Scopus, Ovid MEDLINE, ScienceDirect, Web of Science and EMBASE. Reference lists of related systematic reviews and included studies were manually searched so as to identify additional eligible studies. The complete search strategy was listed in Supplemental File 1.

Studies written in English were included if they fulfilled the following criteria: 1) the target population was old adults aged at least 60 years old; 2) the index test was designed to diagnosis the common health problems (e.g., AF, COPD, falls, diabetes or PD); 3) at least one e-device (including smartphone, smartwatch/wristband and wearable device) was applied in the index test; 4) the index test was suitable for the old adults to perform independently at home; 5) acceptable reference standard was used to judge the diagnostic accuracy; 6) the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) or receiver operating characteristic curves were reported particularly.

Studies were excluded if 1) the participants aged below 60; 2) household unavailable e-devices were applied in the index test (e.g., stationary gait analysis system); 3) the index test was too complicated for the old adults to perform (e.g., excessively complex wearable system or complex questionnaires); 4) no acceptable reference standard was mentioned.

All identified articles were exported to EndNote version 20 where duplications were removed automatically. The remaining articles were screened first according to the titles and abstracts. After that, an assessment with full articles was carried out for eligibility. Studies with inappropriate participants, outcomes and index tests were excluded. At least two reviewers (YY and LZ) were involved in the screening process of titles, abstracts and full papers independently. Any disagreements were resolved by a third reviewer (YC). The number of studies included and excluded at each step and the reasons for exclusion were listed according to the PRISMA diagram (Fig. 1).

Two independent reviewers extract the data from the identified studies according to a standardized data extraction sheet. Again, a third reviewer was involved when any disagreements occurred. The main information included the study characteristics (type of study, year of publication), participant characteristics (number, age, baseline condition), details of index test (e-devices included, method of application, verified setting), acceptable reference standard. Necessary data to calculate the number of true positives, false positives, true negatives and false negatives were extracted additionally. If these data were not reported directly, a manual calculation was conducted according to the value of diagnostic test sensitivity, specificity and the prevalence.

The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was used to assess the risk of bias and concerns about the applicability of study findings to the review question.21 The tool comprised four domains: patient selection, index test, reference standard, and flow and timing. Two independent reviewers did the quality assessment, with any disagreements resolved by referral to a third reviewer.

All the included studies were divided into the following four categories according to the different signals detected: physical activity (PA), vital signs (VS), electrocardiography (ECG) and other.16 Multiple signals or other signals (with the exception of the mentioned three signals) were detected in the study were considered to be included in the ‘other’ category. A meta-analysis was conducted separately according to different categories if the number of included studies were sufficient (> 10) in each category. However, articles classified as “other” would only be analysis by a narrative synthesis due to the limited meaning to analysis the pooled diagnostic accuracy with every article aiming at different health problems.

Heterogeneity was statistically quantified by Q and I-squared (I2). Heterogeneity was defined as low, moderate, and high when the value of I2 was over 25%, 50%, and 75%, respectively.22 When the value of I2 was > 50%, statistical investigation would be conducted in the form of meta-regression.23 Meanwhile, the potential cause of heterogeneity would be explained according to the results of meta-regression and summary receiver operating characteristic (SROC) curves. The meta-regression incorporated four subgroup analyses, namely e-devices included, study design, number of participants, and age distribution. Publication bias was examined by Deek's funnel plot asymmetry test.24 All p-values < 0.05 was considered to be statistically significant.

The pooled sensitivity, specificity, likelihood ratio positive (LR+), likelihood ratio negative (LR-), and diagnostic odds ratio (DOR) with 95% confidence interval (CI) was of each classification were calculated by the bivariate mixed-effects regression model.25,26 A LR+ greater than 10 and an LR- less than 0.1 could stand for strong diagnostic evidence to rule in or rule out diagnoses respectively.27 The DOR was defined as the odds of positive test results in patients with potential health problems compared with the odd of positive results in those without the disease.28 Meanwhile, the SROC curve and corresponding area under the curve (AUC) was constructed with 95% confidence regions for summarizing overall test performance.

STATA Version 15.1 software (StataCorp, TX, USA) was used for meta-analysis and Review Manager (version 5.3, Copenhagen: The Nordic Cochrane centre, The Cochrane Collaboration, 2014) was used for quality assessment.29

A total of 2790 records were identified through database searching, and 2 additional records were added by manual search. 468 duplicate articles were automatically removed before formal screening. After that, a screening according to the title and abstract was carried out, where 1676 results were excluded. Eligibility assessment were conducted by reviewing the full articles of the remaining studies. Finally, 31 studies met the eligibility criteria with 24 studies included in meta-analysis. 7 studies were included in narrative synthesis for the limited number of studies (<10) in the corresponding category. The main reasons for exclusion in each step were listed in the PRISMA diagram (Fig. 1). According to the different signals detected in each study, 13 studies were included in the ‘VS’, 11 studies were included in the ‘ECG’, 3 studies were included in the ‘PA’, and 4 studies were included in the ‘other’.

In the ‘VS’ group, a total of 4844 participants with average age at 70.4 (range from 63 to 78) were included. In the ‘ECG’ group, 2241 subjects with average age at 69.2 (range from 64 to 78) in total were included. In addition, 91 and 650 participants with average age at 67.6 (range from 62.6 to 70.3) and 70.4 (range from 63.5 to 78) were involved in the ‘PA’ and ‘other’ group, respectively.

The index tests were classified by the type of e-devices used in the corresponding study. Four types of e-devices were determined after reading the full texts, namely smartwatch/wristband, smartphone, wearable devices and smartphone combined with additional equipment.30,31 The overall usage of e-devices was showed in Fig. 2.

The included index tests aimed at diagnosing the following health problems: AF, tremor, heart failure, discriminate arrhythmia, acute exacerbations of COPD, depression, falls, obstructive sleep apnea, diabetes, and ulnar artery patency. Among them, 20 studies tested the diagnostic ability of e-device for AF, which accounted for the majority. And 2 studies aimed at the diagnostic accuracy for tremor. For each of the remaining health problems, there was one corresponding study carrying out the works for the diagnostic accuracy of e-device. The overall characteristics of the included studies were summarized in Table 1.

The methodological quality of the studies assessed with the QUADAS-2 tool was summarized in Supplemental File 2. The overall risk of bias and applicability concerns in the index test, reference standard, and flow and timing domains of QUADAS-2 were considered to be low. In the patient selection domain, 9 studies that enrolled consecutive sample of patients and avoided a case-control design were considered to be at high risk of bias.32, 33, 34, 35, 36, 37, 38, 39, 40 In the index test domain, one study was judged as high risk because the knowledge of diagnostic results from the reference standard remained unclear when carrying out the index test.33 In the reference standard domain, two studies were at high risk for the adopted reference standard might not likely to correctly classify the target condition.34,41 In the flow and timing domain, also two studies were considered to be at high risk because it was not mentioned in the articles whether all subjects accepted the same reference standard.33,42