{"id":"31b2f0f3-e811-42ad-94b5-e8b485755d3a","arxiv_id":"1908.06170","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":0.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of remote cardiovascular monitoring concludes that the field needs longitudinal sensors, continuous modeling, and personalized interpretable machine learning.","lead":"This survey reviews wearable sensors and machine learning analytics for monitoring heart disease outside hospitals. It argues that progress requires longitudinal sensing, continuous analytics, and personalized, interpretable models.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The motivating agenda rests on the untested assumption that null telemonitoring trials (Tele-HF, BEAT-HF) failed because sensing/analytics were inadequate; richer data may not change outcomes if failures stem from engagement, workflow, or intervention design.","rationale":"The reader's weakest assumption—that prior telemonitoring null results are due to technological limitations—is indeed the most load-bearing unproven step. The paper's three-needs claim is broad enough to stand on descriptive survey content alone, but the authors explicitly use Tele-HF and BEAT-HF failures to argue that more biomedical signals and longitudinal analytics are needed. That argument requires the failures to be attributable to inadequate sensing/analytics rather than to engagement, care-process, or fundamental monitoring limitations. The reader's proposed condition—that the paper should address alternative interpretations and make selection criteria explicit—is proportionate. A meta-regression of existing trials is a concrete way to test the assumption; if no richness-outcome gradient exists, the motivating interpretation should be downgraded. I therefore see no reason to move the reader's conditional verdict.","tokens_in":30603,"tokens_out":3451,"duration_ms":40845,"concrete_test":"Perform a meta-regression of published heart-failure remote-monitoring RCTs (at minimum Tele-HF [17], BEAT-HF [75], TIM-HF, and WHARF), coding each intervention for sensing richness (self-report only, automated standard vitals, added biosignals/sensors) and analytic sophistication (threshold alerts vs. statistical or machine-learning risk models). Test whether the log-odds reduction in all-cause readmission or mortality increases monotonically with these codes. If no positive gradient is found, the inference that more signals and better analytics would overcome prior null results is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central framing—three primary needs in smart health for cardiovascular disorders—is supported in part by interpreting null telemonitoring trials as evidence that current sensing and analytics are inadequate. Section 1 says Tele-HF and BEAT-HF 'failed at preventing adverse events... suggesting that further exploration of additional biomedical signals are needed.' Section 4.1.3 extends this: with 'more signals captured, and techniques that can better account for varied time-domain aspects of analytics, it is possible that better just-in-time alerts can be generated.' This is a causal counterfactual that the survey does not establish. Tele-HF tested daily telephone-based self-reports [17]; BEAT-HF tested automated transmission of weight, blood pressure, and heart rate with nurse follow-up [75]. Their null results are compatible with non-technological mechanisms: low patient engagement, alert fatigue, lack of a linked treatment pathway, or the possibility that readmissions are not avertable by outpatient monitoring at the tested intensity. The survey cites no evidence that richer signals or more sophisticated analytics change clinical outcomes; it even notes that mSToPS increased AFib diagnosis and anticoagulation but also increased healthcare utilization, with ultimate clinical impact still unknown [97]. Because the survey uses these trials to motivate the three needs, the agenda's force depends on this unverified assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a survey of remote sensing and machine-learning analytics for cardiovascular disorders, using heart failure, coronary artery disease, stroke, and hypertension as case studies. It organizes the literature into three \"primary needs\": sensing that tracks longitudinal signs and symptoms despite noisy or missing data; analytics that model continuous longitudinal data for risk prediction and disease progression; and interpretable, personalized machine learning for shared clinical decision making. After reviewing acoustic, electrical, blood-flow, fluid-retention, and activity sensors, continuous-capture analytics, risk models, and interpretability methods, the paper concludes that addressing the three needs will enable improved remote clinical decision support for cardiovascular patients.","tokens_in":30856,"tokens_out":4571,"duration_ms":42773,"significance":"The survey's main value is organizational: it brings a broad and heterogeneous body of sensing and analytics literature under a single taxonomy and connects each technology category to specific clinical signs and symptoms. It also gives explicit credit to prior work and is generally careful to distinguish demonstrated technical performance from clinical benefit, notably for mSToPS and for edema sensors. As a survey, it makes no new empirical claims and offers no machine-checked artifacts. The conceptual contribution, the three-needs framework, is plausible and potentially useful as a research agenda, but its evidential basis is weaker than the presentation suggests: the motivating interpretation of Tele-HF and BEAT-HF is an unsupported causal counterfactual, and several trial attributions are incorrect. With revisions that sharpen the epistemic claims and correct the citations, the survey could be a useful reference for the smart-health community.","major_comments":[{"comment":"The narrative that Tele-HF and BEAT-HF failed because sensing and analytics were inadequate is a causal counterfactual that the manuscript does not establish. Both trials tested specific interventions (daily telephone self-report; automated transmission of weight, blood pressure, and heart rate with nurse telephone follow-up), and their null results are compatible with poor engagement, alert fatigue, absence of a linked treatment pathway, or limited avertability of readmissions. The survey cites no evidence that richer signals or more sophisticated analytics change clinical outcomes; mSToPS, which the survey itself describes, increased AFib diagnosis and anticoagulation but also increased healthcare utilization with unknown clinical impact. Please either reframe the three needs as a hypothesis or opportunity rather than a conclusion drawn from these trials, or cite direct evidence that data richness or analytic sophistication causally improves outcomes.","section":"Section 1 and Section 4.1.3"},{"comment":"The sentence beginning \"Ong et al. in the Beat-HF study tried to use some machine learning techniques\" misattributes the machine-learning analysis. Reference [84] is a separate secondary analysis by Pourhomayoun et al.; the Beat-HF trial report [75] did not itself evaluate machine-learning models. Please correct the attribution and distinguish the primary trial result from post-hoc analytics.","section":"Section 4.1.3"},{"comment":"The sentence \"In Tele-HF, Krumholz et al. found that a self-report telemonitoring system was not able to reduce readmissions... [48]\" misattributes the null trial result. The primary Tele-HF outcome is reported in Chaudhry et al. [17]; reference [48] is a secondary analysis of non-clinical predictors of readmission. Please correct the citation and keep primary and secondary analyses distinct, since the survey's motivating argument depends on accurate reporting of what these trials actually tested.","section":"Section 5.1.2"}],"minor_comments":[{"comment":"\"MIT-BHI arrhythmia database\" should be \"MIT-BIH arrhythmia database.\"","section":"Section 4.1.2"},{"comment":"The phrase \"q-minima\" would be clearer as \"QRS minima\" or \"Q-wave minima.\"","section":"Section 3.1.2"},{"comment":"\"degredation\" should be \"degradation.\"","section":"Section 4.1.3"},{"comment":"The author name \"BOBAK J. MORTAZA VI,Texas A&M University\" appears with a missing space and comma placement issue; it should be \"BOBAK J. MORTAZAVI, Texas A&M University.\"","section":"Authors' block"},{"comment":"The mSToPS trial is described as ongoing; please clarify that the cited results are from the initial phase and that the three-year follow-up for clinical outcomes is still pending.","section":"Section 4.1.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal. The authors' self-citations are used as examples of established work rather than as circular support, so I do not see a novelty-disclosure concern. The main issue is epistemic: the survey needs to distinguish the authors' research agenda from evidence-based conclusions, and the trial misattributions should be corrected before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this is a competent, useful survey of remote sensing and analytics for cardiovascular disease, and the three-needs framing (longitudinal sensing, continuous analytics, personalized interpretable ML) is a reasonable way to organize the field. But the paper leans on an unverified assumption to motivate that framing: that the null results of Tele-HF and Beat-HF were due to inadequate technology, implying richer signals would change outcomes. That inference is not established.\n\nWhat it does well: the survey is broad and clinically informed. It covers sensing modalities, continuous data analysis, and interpretable models, and it integrates clinical context (HF, CAD, stroke, HTN) rather than treating algorithms in a vacuum. The summaries of individual papers are mostly accurate, and the identification of gaps—missing data, sensor burden, transfer learning, dynamic risk models, interpretability—is sensible. The self-citations are relevant and not circular; this is a survey, so no new result is expected.\n\nThe soft spot is the motivating claim. The paper says Tele-HF and Beat-HF 'failed' and that 'further exploration of additional biomedical signals are needed.' That's a causal counterfactual. Those trials had specific designs, and their null results are compatible with low engagement, alert fatigue, lack of a treatment pathway, or simply that the intervention intensity was too low. The survey even notes that mSToPS increased AFib diagnosis and anticoagulation but also increased utilization with unknown clinical impact—so it's aware of the complexity, but it doesn't wrestle with the possibility that more data won't help. The three-needs agenda is weaker if the failure is not primarily technological. Also, the survey doesn't state its selection criteria for included papers, which is a minor but fixable omission.\n\nBottom line: this is a solid review that will be a helpful entry point for researchers coming into smart health for cardiovascular disorders. It deserves a serious referee, but the authors should be asked to address the alternative explanations for the trial nulls and either support the technological-limitation claim or soften it. I'd bring it to a reading group to argue about that exact issue.","headline":"A well-organized survey of cardiovascular remote monitoring, but its motivating claim that null telemonitoring trials failed for technological reasons is an unverified leap.","tokens_in":31318,"tokens_out":2380,"would_cite":true,"duration_ms":25928,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Remote monitoring for cardiovascular disorders has failed in trials so far, but this survey argues the failures came from thin sensing and analytics, not from the idea itself.","keywords":["cardiovascular disease","longitudinal monitoring","smart health","wearable sensors","machine learning","heart failure","remote patient monitoring","interpretability"],"falsifier":"A randomized telemonitoring trial for heart failure that fully implements all three needs, with continuous multimodal sensing, longitudinal models that update risk, and interpretable personalized alerts, and then shows no reduction in readmissions or mortality despite high adherence and clean data would refute the survey's diagnosis. A cheaper test would re-analyze Tele-HF and Beat-HF data to determine whether patients who were fully adherent and received timely alerts still showed no benefit; if they did not, data richness was not the limiting factor.","tokens_in":30455,"feed_emoji":"🫀","tokens_out":6463,"duration_ms":60283,"temperature":0.7,"pith_summary":"This survey argues that remote monitoring for cardiovascular disorders has so far failed not because the concept is wrong, but because the technology and analytics have been too thin. It organizes the field around three needs: sensing that tracks longitudinal trends through noisy, infrequent, or missing measurements; analytics that model continuously captured data for risk prediction and disease progression; and machine learning that is personalized and interpretable enough to support shared clinical decisions. As a survey, its contribution is a synthesis and an agenda rather than a new measurement: it assembles the sensor modalities and analytic techniques that already exist, identifies where each falls short, and points to the integrations that would make longitudinal cardiac monitoring clinically effective. The stakes are practical, since cardiovascular disorders account for nearly one in three deaths in the United States and most decisions are made from brief acute-care encounters.","feed_headline":"Remote heart monitoring failed on weak data, not on the idea","feed_subtitle":"Richer sensors, longitudinal analytics, and interpretable machine learning could make remote cardiac care work.","key_machinery":"The organizing device is a three-need framework that acts as a pipeline: longitudinal sensing to continual analytics to personalized, interpretable modeling. Each sensor category, including acoustic and vital-sign capture, electrical measurements, blood-flow and cuff-less blood pressure, fluid retention and edema tracking, and physical activity and posture, is assessed against the first need; time-series and deep-learning methods are assessed against the second; and risk models such as CHA2DS2-VASc, HAS-BLED, QStroke, and the Pooled Cohort Equations are assessed against the third. The framework does the argumentative work by converting scattered sensor papers and trial null results into a single diagnosis of what is missing and where to build.","core_discovery":"The central claim is stated in the abstract and repeated in the discussion: new smart health technologies for heart failure, coronary artery disease, and stroke need to satisfy three conditions at once. First, sensors must capture longitudinal trends in signs and symptoms even when measurements are infrequent, noisy, or missing, because patients will not wear every sensor all the time and context changes measurement quality. Second, analytic techniques must model data in a longitudinal, continual fashion, moving beyond anomaly detection and fixed-window classification toward tracking disease progression and updating risk over time. Third, the resulting machine learning must be personalized and interpretable, so patients and clinicians can act on predictions, update them as treatment changes, and understand the drivers of risk. The paper reads the failed Tele-HF and Beat-HF telemonitoring trials as evidence that point measurements and simple alerts are insufficient, and surveys existing acoustic, electrical, blood-flow, fluid-retention, and activity-sensing work to argue that the missing pieces are now being developed.","pith_inferences":["If the three-needs framing is correct, the natural unit of clinical validation shifts from a single wearable device to an integrated sensing-plus-analytics system, and future trials should measure adherence, data quality, and alert response as mediators rather than only the final event rate.","The framework implies a transfer-learning research program: because patients will not wear every sensor continuously, models that can estimate missing modalities from worn ones, such as inferring blood pressure from pulse arrival time or activity context, become central to longitudinal monitoring.","The same pipeline could extend to other chronic conditions characterized by slow decompensation, such as diabetes or chronic obstructive pulmonary disease, following the paper's own analogy from human activity recognition to cardiac monitoring.","A relatively cheap test of the survey's premise would be a retrospective simulation on Tele-HF and Beat-HF data that imputes missing daily weights and activity levels and checks whether richer longitudinal features would have produced earlier or more accurate decompensation alerts."],"forward_implications":["If the three needs are accepted, remote cardiac monitors should be evaluated on how well they track trends through missing and noisy data, not on occasional point measurements.","Risk scores built on sparse history, such as CHA2DS2-VASc, could be enriched with temporal data such as atrial fibrillation burden, potentially sharpening decisions about anticoagulation.","The framework implies that integrating multiple sensing modalities into a single platform, such as wrist and chest devices, is a prerequisite for longitudinal monitoring because patients will not wear many separate sensors.","Machine learning for telemonitoring must be interpretable and personalized, since clinicians cannot act on outputs they cannot explain or adapt to a specific patient.","The survey's reading of the Tele-HF and Beat-HF trials predicts that a telemonitoring system satisfying all three needs could reduce heart failure readmissions even though earlier simple alerts did not."],"supporting_citations":[{"why":"Tele-HF trial, whose null result motivates the need for richer sensing and analytics.","marker":"[17]"},{"why":"Analysis of Tele-HF data showing self-reported information can improve readmission prediction, grounding the analytics opportunity.","marker":"[48]"},{"why":"Beat-HF trial, where automated vital-sign telemonitoring also failed to reduce readmissions, supporting the need for additional signals.","marker":"[75]"},{"why":"Survey of telemedicine in heart failure used to argue that outcomes depend on personalization to the particular patient.","marker":"[5]"},{"why":"mSToPS trial, where home ECG monitoring detected more atrial fibrillation than routine care, evidence for continuous electrical capture.","marker":"[97]"},{"why":"ACC/AHA risk assessment guideline with the Pooled Cohort Equations, a baseline risk model that longitudinal data could enrich.","marker":"[31]"},{"why":"CHA2DS2-VASc stroke risk model, a sparse model that temporal data such as atrial fibrillation burden could improve.","marker":"[56]"},{"why":"TOPCAT trial, whose regional and subgroup variation in treatment response motivates personalized, interpretable modeling.","marker":"[83]"}],"fun_headline_variants":["Heart monitoring’s real problem: weak data, not weak tech","Three fixes could make remote cardiac monitoring work","Remote heart care needs longitudinal data, not just alerts","Interpretable ML and messy data: keys to cardiac monitoring","Why heart telemonitoring failed: missing longitudinal analytics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the null results of earlier telemonitoring trials were caused by inadequate sensing and analytics, rather than by something else about remote monitoring, so that richer sensors and better models can actually reduce adverse events.","fun_headline_variants_meta":{"raw":{"variants":["Heart monitoring’s real problem: weak data, not weak tech","Three fixes could make remote cardiac monitoring work","Remote heart care needs longitudinal data, not just alerts","Interpretable ML and messy data: keys to cardiac monitoring","Why heart telemonitoring failed: missing longitudinal analytics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000191,"raw_usage":{"total_tokens":1371,"prompt_tokens":998,"completion_tokens":373,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":614,"completion_tokens_details":{"reasoning_tokens":295}},"tokens_in":614,"tokens_out":373,"duration_ms":3976,"temperature":1.0,"reasoning_tokens":295,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:54:31.869157+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A randomized telemonitoring trial for heart failure that fully implements all three needs, with continuous multimodal sensing, longitudinal models that update risk, and interpretable personalized alerts, and then shows no reduction in readmissions or mortality despite high adherence and clean data would refute the survey's diagnosis. A cheaper test would re-analyze Tele-HF and Beat-HF data to determine whether patients who were fully adherent and received timely alerts still showed no benefit; if they did not, data richness was not the limiting factor.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Tele-HF trial, whose null result motivates the need for richer sensing and analytics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Beat-HF trial, where automated vital-sign telemonitoring also failed to reduce readmissions, supporting the need for additional signals."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Survey of telemedicine in heart failure used to argue that outcomes depend on personalization to the particular patient."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"TOPCAT trial, whose regional and subgroup variation in treatment response motivates personalized, interpretable modeling."}],"review_version":1}