The Assessment of the Readiness of Molecular Biomarker-Based Mobile Health Technologies for Healthcare Applications

Mobile health technologies to detect physiological and simple-analyte biomarkers have been explored for the improvement and cost-reduction of healthcare services, some of which have been endorsed by the US FDA. Advancements in the investigations of non-invasive and minimally-invasive molecular biomarkers and biomarker candidates and the development of portable biomarker detection technologies have fuelled great interests in these new technologies for mhealth applications. But apart from the development of more portable biomarker detection technologies, key questions need to be answered and resolved regarding to the relevance, coverage, and performance of these technologies and the big data management issues arising from their wide spread applications. In this work, we analyzed the newly emerging portable biomarker detection technologies, the 664 non-invasive molecular biomarkers and the 592 potential minimally-invasive blood molecular biomarkers, focusing on their detection capability, affordability, relevance, and coverage. Our analysis suggests that a substantial percentage of these biomarkers together with the new technologies can be potentially used for a variety of disease conditions in mhealth applications. We further propose a new strategy for reducing the workload in the processing and analysis of the big data arising from widespread use of mhealth products, and discuss potential issues of implementing this strategy.

There have been intensifying efforts to explore mobile health (mhealth) technologies for delivering healthcare at reduced costs and for facilitating more precise and personalized medicine [1][2][3] which have led to 73 apps endorsed (examples in Table 1, a complete list in Supplementary Table S1) and additional ones reviewed 1 by the US Food and Drug Administration (FDA) for self-diagnosing acute diseases and monitoring chronic conditions 1 based on such physiological biomarkers as body temperature and brainwave 4,5 , and such simple-analyte biomarkers as glucose and urine protein contents 4,5 .
Although these physiological and simple-analyte biomarkers cover many disease conditions, their coverage is substantially limited for such prevalent diseases as cancers, infectious, respiratory, digestive, endocrine and nervous system diseases, as indicated by the disease-coverage profiles of the 73 FDA endorsed, and 94 physiological and simple-analyte biomarker candidates described in the literatures (Fig. 1, Table 1 and 2, Supplementary Table S1 and S2). Apart from the development of more portable biomarker detection technologies, additional biomarkers are needed for fulfilling the tasks of mhealth technologies as efficient and effective means for providing wider coverage of healthcare and personalized treatments at reduced costs [1][2][3] .
Some genetic, proteomic and metabolomic molecular biomarkers have been clinically used and many more such molecular biomarker candidates (hitherto also tentatively named biomarkers) have been discovered for diagnosing and monitoring diseases, directing treatments and predicting patient responses [6][7][8] . Of immediate 1 Shenzhen Kivita Innovative Drug Discovery Institute, and the Ministry-Province Jointly Constructed Base for State Key Lab-Shenzhen Key Laboratory of Chemical Biology, the Graduate School at Shenzhen, Tsinghua University, Shenzhen, P. R. China. 2 Department of Pharmacy, and Center for Computational Science and Engineering, National University of Singapore, 117543 Singapore. 3 NUS Graduate School for Integrative Sciences and Engineering, National University of Singapore, 117456, Singapore. 4 Computation and Systems Biology, Singapore-MIT Alliance, National University of Singapore, Singapore. Correspondence and requests for materials should be addressed to Y.Y.J. (email: Jiangyy@ sz.tsinghua.edu.cn) or Y.Z.C. (email: phacyz@nus.edu.sg) relevance to mhealth are the hundreds of literature-reported non-invasive and minimally-invasive diagnostic, prognostic and theragnotic molecular biomarkers from such non-invasive sources as urine, breath, saliva, tear, feces, sputum and oral mucosa samples (Examples in Table 3 and complete list in Supplementary Table S3) and from such minimally-invasive sources as finger-prick (the list of serum biomarkers potentially detectable from finger-prick is in Supplementary Table S4), which significantly expand the disease coverage as indicated by the disease-coverage profiles of the 664 (27 clinical trial) non-invasive and 592 serum (69 clinical trial or use) molecular biomarkers with respect to those of 73 FDA endorsed apps and 94 physiological and simple-analyte biomarkers (Fig. 1). Many biomarkers are detectable by the new biomarker-detection technologies that become increasingly portable, faster, user-friendly, inexpensive and accurate 9,10,11 , some of which have been explored for potential mhealth applications 9,[12][13][14][15] .
From the investigations and opinions described in the literatures listed in Supplementary Table S3, there are good reasons to speculate the readiness of some of these technologies for mhealth applications. But before the acceptance and widespread utilization of these technologies, several key questions need to be answered or resolved. Apart from the development of more portable biomarker detection technologies, an important question is whether the new portable biomarker detection technologies are sufficiently sensitive, fast, convenient and inexpensive for biomarker detection in the typical mhealth settings (low sample volume and biomarker concentrations). Another question is whether the discovered and investigative molecular biomarkers extracted from the non-invasive and minimally invasive sources are relevant to mhealth applications in terms of the detection accuracies and the coverage of disease conditions and patient populations. The third is how to resolve the different readings generated from different mhealth devices and variations in individual operations. The fourth is how to manage the heavy workload in processing and analysing the big data arising from widespread use of mhealth devices.
Here, we address some of these questions by analysing (1)  test time, and costs with respect to experimentally-determined biomarker levels in patients and the detection limits, and (2) the disease coverage, patient populations, and the diagnostic, prognostic, and theragnostic sensitivity and specificity of the literature-reported non-invasive and minimally-invasive finger-prick molecular biomarkers for mhealth applications with respect to the detection limits of the new detection technologies. We also discuss the feasibility and practical issues of adopting a new strategy for reducing the heavy workload of mhealth data processing by automated electronic pre-screening of the big biomarker screening data.

Literature Search
The detailed information of 73 mhealth apps endorsed by the US FDA was obtained by manually checking the descriptions of the apps listed in FDA 510(k) medical device database 16 . The physiological and molecular biomarkers were obtained by the comprehensive literature search of the Pubmed database by using the combination of the keyword "biomarker" together with one of the keywords of "clinical", "patient", "disease", "drug", and specific disease names such as "cancer", "inflammation" and "hypertension". We also searched and evaluated biomarker review papers from reputable journals by using the combination of the keywords "biomarker" and "review", with the cited original articles checked to collect detailed information about the discussed biomarker, such as the name, source, specific disease and function, specificity and sensitivity of the biomarker. The detailed information of these 254 evaluated review and research papers are listed in Supplementary Table S6. Additional sources such as the abstracts Mental, Behavioral and Neurodevelopmental disorders, G00-G99: nervous system disorders, H00-H59: eye and adnexa diseases, H60-H95: Diseases of the ear and mastoid process, I00-I99: circulatory system disorders, J00-J99: respiratory system disorders, K00-K95: digestive system disorders, L00-L99: skin and subcutaneous tissue disorders, M00-M99: musculoskeletal system and connective tissue disorders, N00-N99: genitourinary system disorders, O00-O9A: Pregnancy, childbirth and the puerperium, P00-P96: conditions originating in the perinatal period, Q00-Q99: Congenital malformations, deformations and chromosomal abnormalities, R00-R99: conditions not elsewhere classified, S00-T88: Injury, poisoning and certain other consequences of external causes, V00-Y99: External causes of morbidity, Z00-Z99: Factors influencing health status and contact with health services of the American society of clinical oncology were also systematically searched, with 658 biomarker conference abstracts in 1995-2013 extracted and evaluated by data mining and manual curation. Non-invasive biomarkers were selected if they were detected in non-invasive tissues such as urine, breath, saliva, tear, feces, sputum and oral mucosa samples. The information of disease conditions was searched from the websites of professional medical associations such as WHO 17 and American Cancer Society 18 , and such additional sources as reputable books and review articles, using combinations of keywords such as the disease name and "prevalence" or "incidence". These biomarkers were organized based on their international classification ICD-10 codes 19 and were displayed with respect to these codes in a tree graph by using the automatic tree generator module in iTOL 20 .
The performance of the biomarkers in diagnosing, prognosing or theragnosing specific conditions has been statistically measured by sensitivity (the proportion of the condition-positive samples that are correctly identified as positive) and specificity (the proportion of the condition-negative samples that are correctly identified as negative) 21 . Wherever reported in the literature, these statistical performance measures were recorded. Apart from the collection of the biomarker detection technologies described in our searched biomarker literatures, additional literature search was conducted for searching biomarker detection technologies of potential mhealth applications by using the keyword "biomarker" in combination with one of the keywords "detection", "detector", "device", "technology", "technique" and "assay". These detection technologies were analysed for selecting those with potential mhealth applications based on their detection performance, portability, detection time, cost and ease of use.

New technologies for detecting non-invasive molecular biomarkers and their relevance to mhealth
The new biomarker-detection technologies combined with mobile phone or the equivalent imaging devices have been explored for detecting at least 23 molecular biomarkers including 11 non-invasive ones (Table 4). These new technologies can be categorized into four groups: (1) paper-based and mobile phone enabled, (2) paper-based, (3) mobile-phone enabled, and (4) the other point of care technologies. The first group of technologies combines innovative paper-based microfluidic analytical technologies with mobile phone enabled automated image processing tools, which are most relevant to mhealth applications because of the very low cost (~$2.60+ cost plus mobile phone), increasingly enhanced detection sensitivity (0.3-60 ng/mL, 0.13-21.3 μ g/mL and 0.81-2000 ng/mL for small molecule, peptide and protein biomarkers respectively), low sample volumes (0.5-25 μ L), short detection time (15-60 mins), and the convenient biomarker processing (mobile phone-based) capabilities. The recently developed paper-based microfluidic analytical technologies include paper-based enzyme-linked immunosorbent assays (P-ELISA) 9,22 , paper lateral flow immunoassays (P-LFIAs) 12,23 , and paper-based Au-nanoprobes 22  are integrated with or coupled to mobile phones equipped with the colorimetric algorithms 22 and the applications for immediate data processing of the detection results without referring to peripheral equipment for read-out and analysis 9 . The second group of technologies primarily employ innovative P-ELISA in combination with a scanner, printer or digital camera based image-processing facility to achieve a detection sensitivity as high as 33.7 fg/mL 24 and 18 pM/mL 25 for detecting peptide and protein biomarker respectively. The imaging processing component of these technologies may be potentially replaced by mobile phone-based ones for potential mhealth applications. The third group of technologies integrates mobile phone imaging processing tools with newly developed disposable microfluidic chip 26 , opto-acoustic immunoassay 27 , microfluidic capillary array equipped with optical signal amplifier 28 , microtiterplate based ELISA 29 and other technologies. These technologies achieve detection sensitivity up to the level of 60-300 pg/mL for protein biomarkers 29,30 . Although their costs are more suitable for point of care (POC) rather than mhealth applications, the innovative design may be potentially implemented into paper-based platforms for more extensive mhealth applications. A new POC technology in the fourth group, the negative-pressure-driven microfluidic chip magnetic bead based ELISA, is capable of detecting a small molecule biomarker at sensitivity level of 0.3 ng/mL 31,32 . If implemented into paper-based and mobile phone-enabled platforms, this technology may potentially find wider applications for detecting small molecule biomarkers in mhealth.
Overall, 12 or 52.2% of the 23 tested molecular biomarkers are detectable by these new technologies at low concentrations (0.3-810 pg/mL and 4-50 ng/mL for 8 and 4 biomarkers respectively). Although the detectable concentrations of these 23 biomarkers are roughly 10-fold higher than those of the conventional technologies 24 , seven of them are nonetheless within the lower detection limit of the new technologies for non-invasive detection 24,27 . Of the eight biomarkers with available patient data, only two biomarkers in the corresponding non-invasive source are outside the detection limit of the new technologies. Moreover, 64.3% of these biomarkers are detectable at  Continued significantly lower sample volumes (0.5-12 μ L) and shorter time (10-60 min) than the volumes (100-300 μ L) 13,25 and durations (up to 4h) 24 of the conventional technologies. The costs of these detection devices are ~$300-$600 US dollars 33 . The per-test costs are in the range of 0.01-190. Therefore, the new technologies are fairly sensitive, efficient, and inexpensive for detecting a substantial percentage of the tested non-invasive biomarkers, and there is high likelihood that they can be applied for detecting other non-invasive biomarkers in mhealth applications.

The non-invasive molecular biomarkers and their relevance to mhealth
Analysis of the 664 literature-reported non-invasive molecular biomarkers (examples in Table 5 and a complete list in Supplementary Table S5) Tables 3 and 5 and a complete list in Supplementary Table S3, S5) Their performances have been typically measured by sensitivities (the rates for positive identification of disease conditions) and specificities (the rates for correct classification of the negatives). The sensitivities and specificities of the majority of these biomarkers are ≥ 85% and ≥ 80% for diagnosis, and ≥ 80% and ≥ 80% for prognosis respectively, which are roughly at the ≥ 90% sensitivity and ≥ 90% specificity levels of the good biomarkers 21 . Therefore, a substantial percentage of these non-invasive biomarkers are expected to be potentially useful for pre-screening patients in need of further evaluations in mhealth applications.
The utility of these biomarkers for mhealth applications also depends on whether they are detectable by the new detection technologies, i.e., whether the levels of these biomarkers in the non-invasive sources from the patients are within the detection range of the new detection technologies. We searched from the literatures the corresponding biomarker levels for 35 diseases (Supplementary Table S5, examples in Table 5) and compared them to the detection limits of the new technologies. Our analysis showed that 26 (or 74.3%) of the 35 disease conditions with searchable information, including 8 disease conditions with large patient populations, have one or more biomarker detectable by the new technologies ( Table 5), suggesting that a substantial percentage of the disease conditions including those with large patient populations may be partly covered by the new technologies.

The potential of the minimally invasive finger-prick biomarker technologies for mhealth applications
The minimally invasive finger-prick biomarker technologies have been developed for POC applications 11 . Because of their improved detection performance 34 , portability 35 and ease of use 36 , and because of their decreased detection time 34 , some of these technologies when combined with smartphone-based processing technologies may find potential mhealth applications. Serum biomarkers are known to be detectable at finger-prick albeit at altered concentrations and thus at re-adjusted detection cut-off values 37,38 . Therefore, one can hypothesize that most of the serum biomarkers of sufficient level of concentrations may be potentially detectable by finger-prick biomarker technologies. The application of these technologies in mhealth significantly expands the coverage of disease conditions because some biomarkers not found in urine are in the serum (e.g. it has been reported that the blood contains the common markers of liver function that are not found in urine 35 ). Our own literature search results showed that the literature-reported serum biomarkers and biomarker candidates cover additional 62 disease conditions beyond those covered by the existing physiological, simple-analyte, and the non-invasive molecular biomarkers and biomarker candidates ( Fig. 1 and Supplementary Table S4).  Moreover, the finger-prick biomarker technologies can potentially have more enhanced capabilities in detecting the biomarkers of low concentrations. The levels of biomarkers in blood are typically more concentrated than those biomarkers collected from the non-invasive urine, breath, saliva, tear, feces, sputum or oral mucosa sources{Song, 2014 #89} {Abdalla, 2012 #115}. For those biomarkers with concentrations in the non-invasive and finger-prick sources below and above the detection limit of the mhealth biomarker technologies respectively, some of them are potentially detectable by using finger-prick biomarker technologies even if they are undetectable by the non-invasive biomarker technologies.
Several new technologies have been developed with potential applications for detecting serum biomarkers from a drop of blood (Table 4). To enable the purification and detection of serum biomarkers, specially designed fluid handling and silver reduction devices have been combined with the ELISA microfluidic chip for simplified biomarker detection, which enables the detection of an HIV biomarker from 1 μ l of unprocessed whole blood in < 15 min 39 . In another design, a microfluidic purification chip was developed for simultaneously capturing multiple biomarkers from blood samples and releasing them into purified buffer for sensing by a silicon nanoribbon detector, which was able to detect two model cancer antigens from a 10 ml sample of whole blood in < 20 min 40 . A micropatterned paper device that combines a filter membrane and a patterned paper chip for achieving blood plasma erythrocyte separation and biomarker detection from the blood from a fingerstick, which is capable of detecting protein biomarkers at ~50 g/L concentrations 35 . Progress has been made in developing plasmonic ELISA for the ultrasensitive detection of disease biomarkers with the naked eye with the ability to detect biomarkers in whole serum at the ultralow concentration of 10 −18 g mL −1 41 .
We have found the reports about the detection of 12 serum biomarkers by means of these new technologies (Table 4). Overall, 5 or 42% of the 12 biomarkers are detectable at concentrations of < 1.5 ng/mL. Considering that many serum biomarker concentrations are higher than those collected from the urine or other non-invasive sources, the relevant technologies may be extended for the detection of a more variety of low concentration biomarkers than those coverable by the non-invasive biomarker technologies. These technologies enable serum biomarker detection mostly at low sample volumes of 1-10 uL and short time of 12-30 min comparable to those of the non-invasive biomarker technologies. The cost of a microtiterplate based ELISA device coupled with a smartphone is < $660 29 . The per test costs of these technologies are in the range of $0.1-34. Three studies reported the sensitivity and specificity of five serum biomarkers, which are in the range of 82-100% (vast majority > 90%) and 78%-100% respectively 38,39,42 . Therefore, these new technologies are fairly sensitive, efficient, and inexpensive for detecting a substantial percentage of the tested serum biomarkers with potential mhealth applications.

Coping with the heavy workload in mhealth: Feasibility of automated electronic prescreening of big mhealth data
There are concerns about the increased workload in processing and analysing the big data arising from widespread use of mhealth devices 1 . On the hand, mhealth devices as digital tools may conveniently facilitate electronic pre-screening of the biomarker readings for filtering potential patients likely in need of further attention and evaluation, which helps to significantly reduce the workload. A digitally-coded biomarker, disease and therapeutic information processing system may be developed for automatically receiving, processing, pre-screening, and dispatching the biomarker readings transmitted from mhealth devices (Fig. 2).
It is feasible to develop such a system using available tools such as the International Classification of Diseases (ICD) codes for defining, studying and managing diseases and treatments 43 , the Systematized nomenclature of medicine for clinical documentation and reporting 44 , the Unified medical language system for biomedical terminology 45 , the Therapeutic target database biomarker and target information and links to the ICD and drug codes 46 , and the Drugbank drug information 47 . Further efforts are needed for additional information refinement and integration, determination and clinical validation of biomarker pre-screening thresholds, and development and  education of testing protocols. There are also potential issues arising from missed detection or misidentification by an electronic system, lack of data security and insufficient regulation standards.

Concluding Remarks
Molecular biomarker-based mobile health technologies have the potential to significantly improve the efficiency and quality of healthcare for a variety disease conditions particularly those with large patient populations that cannot be solely covered by physiological and simple-analyte biomarkers. Some of these biomarkers combined with the new detection technologies are readily applicable for mhealth applications. The increased workload in processing and analyzing high volumes of mhealth data may be efficiently managed by an electronic system that facilitate automatic pre-screening and analysis of the biomarker data for filtering potential patients likely in need of further attention and evaluation.