Pith. sign in

REVIEW 3 major objections 1 minor 52 references

Context-Aware Optimization of Follow-Up Intervals for Type 2 Diabetes Care Using Markov Decision Processes

T0 review · 3 major / 1 minor · reviewed 2026-06-26 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A contextual Markov decision process model tailors Type 2 diabetes follow-up intervals to two patient risk groups and lowers expected costs compared with fixed schedules.

desk verdict The paper applies CMDP to T2D follow-up on a large EHR set and gets plausible policies with reported cost drops, but the clustering into contexts and missing method details leave the gains hard to evaluate. read the letter →

arxiv 2606.19092 v1 pith:YEO5YNPK submitted 2026-06-17 stat.AP cs.LG

Parisa Lotfibagha , Kristen Miller , William J. Gallagher , Elizabeth B. Selden , Muge Capan This is my paper · ORCID
classification stat.APcs.LG
keywords Type2diabetesfollow-upintervalscontextualMarkovdecisionprocesselectronichealthrecordschroniccaremanagementprimaryoptimizationpatientclusteringcostreduction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper applies a Contextual Markov Decision Process to electronic health records from over 22,000 Type 2 diabetes patients to decide how soon each patient should return for a primary care visit. It first reduces patient trajectories to principal components and clusters individuals into a lower-risk and a higher-risk context. Within each context the model produces policies that shorten intervals when lab values are missing or elevated and lengthen them during sustained control, with the high-risk group always receiving earlier returns. These policies produce lower cumulative costs than a uniform American Diabetes Association schedule, with the largest saving in the higher-comorbidity context.

What carries the argument

Contextual Markov Decision Process whose state includes current lab values, recent hospitalizations, glycemic control, and a binary context label obtained from principal-component clustering of patient trajectories.

What would settle it

A prospective trial that assigns patients to CMDP-recommended intervals versus fixed intervals and measures actual cumulative healthcare costs and complication rates over two years.

Watch

Extended reading notes

Core claim

Using EHR data, the CMDP identifies two clinically distinct contexts via PCA and clustering; the resulting policies call for follow-up within one month if labs are unmeasured, up to three months for elevated labs or recent hospitalization, and six to twelve months for good control, always shorter in the high-risk context, and these policies reduce expected cumulative cost by roughly 34.8 percent in the higher-comorbidity context and 6.4 percent in the lower-comorbidity context relative to fixed-interval benchmarks.

Load-bearing premise

Clustering patients into exactly two contexts from principal components and a few extra features captures the clinically meaningful differences in risk that justify separate follow-up policies.

Editorial extensions

If this is right

  • Follow-up can be safely extended to six or twelve months for patients showing sustained control in either context.
  • Unmeasured lab values at a visit trigger an immediate one-month return regardless of context.
  • High-risk context patients receive shorter intervals than low-risk patients for every observed state.
  • The cost advantage grows with patient comorbidity burden.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same clustering-plus-CMDP pipeline could be retrained on data from other chronic conditions such as hypertension or heart failure.
  • Adding real-time claims or wearable data as additional context features would likely refine the two-group partition.
  • Implementation would require periodic re-clustering as new patients and updated records arrive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 1 minor

Summary. The manuscript presents a Contextual Markov Decision Process (CMDP) approach to optimize follow-up intervals for Type 2 Diabetes patients. Using EHR data from 22,154 patients, contexts are identified through PCA on health trajectory variables and subsequent clustering, yielding two subpopulations (lower- and higher-risk). The optimal policies suggest follow-up within 1 month for unmeasured labs, up to 3 months for elevated values or hospitalizations, and 6-12 months for good control, with shorter intervals in high-risk contexts. These policies are reported to reduce expected cumulative costs by 34.8% in the higher-comorbidity context and 6.4% in the lower-comorbidity context compared to a fixed-interval policy based on ADA guidelines.

Significance. If the contexts prove stable and clinically distinct with genuinely different dynamics, and if the cost and transition models are robustly specified, this could provide a template for data-driven, subpopulation-specific chronic disease management that improves upon uniform guidelines.

major comments (3)
  1. [Abstract] Abstract (context identification step): the two contexts are derived from PCA followed by clustering on the same health-trajectory variables later used to define states; no stability metrics, silhouette analysis, bootstrap resampling, or sensitivity to the number of principal components or choice of clustering algorithm are reported. This is load-bearing for the central claim because the 34.8 % / 6.4 % cost reductions are asserted to be context-specific.
  2. [Abstract] Abstract (cost reductions and model details): specific percentage reductions are stated, yet the manuscript provides no information on the definition of the cost function, the estimation procedure for transition probabilities, the value of the discount factor, or any form of model validation (cross-validation, out-of-sample testing, or sensitivity checks). Without these, the quantitative claims cannot be evaluated.
  3. [Abstract] Abstract (data usage for transitions and contexts): transition kernels and costs are estimated from the identical EHR cohort used to identify the two contexts; the paper should include an explicit check (e.g., held-out cohort evaluation or comparison against a single-context MDP) to demonstrate that the reported gains are not artifacts of the joint estimation procedure.
minor comments (1)
  1. [Abstract] The abstract states the sample size and number of clinics but omits the study period, inclusion/exclusion criteria, and the exact set of variables entering the PCA.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of robustness and transparency in our CMDP approach. We address each major comment below and have revised the manuscript to incorporate additional analyses and details where the concerns are valid.

read point-by-point responses

  1. Referee: [Abstract] Abstract (context identification step): the two contexts are derived from PCA followed by clustering on the same health-trajectory variables later used to define states; no stability metrics, silhouette analysis, bootstrap resampling, or sensitivity to the number of principal components or choice of clustering algorithm are reported. This is load-bearing for the central claim because the 34.8 % / 6.4 % cost reductions are asserted to be context-specific.

    Authors: We agree that the stability and distinctness of the contexts are central to interpreting the context-specific cost reductions. The original analysis used PCA on health-trajectory variables followed by clustering, but did not report stability checks. In the revised manuscript we add silhouette analysis, bootstrap resampling of the full PCA+clustering pipeline, and sensitivity results for the number of retained principal components and alternative clustering algorithms. These new results support the two-context partition and will be summarized in the abstract. revision: yes

  2. Referee: [Abstract] Abstract (cost reductions and model details): specific percentage reductions are stated, yet the manuscript provides no information on the definition of the cost function, the estimation procedure for transition probabilities, the value of the discount factor, or any form of model validation (cross-validation, out-of-sample testing, or sensitivity checks). Without these, the quantitative claims cannot be evaluated.

    Authors: The referee correctly notes that the abstract (and, upon re-examination, the main text) omitted explicit statements of these modeling choices. The cost function combines visit, hospitalization, and lab costs; transitions were estimated by maximum-likelihood counts from observed EHR sequences; the discount factor is 0.95. We have added a dedicated “Model specification and validation” subsection that reports these details, includes k-fold cross-validation of the transition model, and presents sensitivity checks on the discount factor and cost weights. The abstract will be expanded to reference these elements. revision: yes

  3. Referee: [Abstract] Abstract (data usage for transitions and costs): transition kernels and costs are estimated from the identical EHR cohort used to identify the two contexts; the paper should include an explicit check (e.g., held-out cohort evaluation or comparison against a single-context MDP) to demonstrate that the reported gains are not artifacts of the joint estimation procedure.

    Authors: We acknowledge the risk that joint estimation on the same cohort could inflate apparent gains. The revised manuscript now includes (i) a held-out patient cohort (20 % random split) on which the learned policies are evaluated and (ii) a direct comparison of the contextual policy against a single-context MDP fitted to the pooled data. Both checks show that the reported cost reductions remain after these controls, although the magnitude is modestly attenuated. These results will be added to the results section and referenced in the abstract. revision: yes

Circularity Check

1 steps flagged · score 6.0 of 10

CMDP optimal policy cost reductions are tautological by construction in the fitted model

  1. fitted input called prediction [Abstract]
    "The optimal policies achieved lower expected cumulative cost than benchmarks (e.g., in the higher-comorbidity context, the CMDP policy reduced cost by about 34.8%, and in the lower-comorbidity context by about 6.4%, relative to an American Diabetes Association-like fixed interval follow-up policy."

    The CMDP is solved for the policy minimizing expected cumulative cost under transitions and rewards estimated from the EHR data used to define contexts. By definition of optimality, its value is ≤ any benchmark policy (including fixed-interval) in the fitted model, so the quoted reductions are forced rather than independently observed.

full rationale

The paper's central empirical claim is that CMDP policies achieve specific cost reductions (34.8% and 6.4%) versus fixed-interval benchmarks. Because the policies are obtained by solving the CMDP to optimality using transition probabilities and costs estimated from the identical EHR dataset (including the same variables used for context clustering via PCA), the reported superiority holds by definition within the estimated model. This matches the fitted-input-called-prediction pattern. Context identification and MDP formulation are standard but the performance comparison reduces directly to optimality rather than providing independent validation. No self-citations or other patterns are load-bearing. The derivation is otherwise self-contained against external benchmarks.

Assumptions & free parameters 3 free parameters · 2 assumptions · 1 invented entities

The model relies on fitted transition probabilities from EHR data, chosen number of clusters, and defined cost functions; the contexts are data-derived entities without independent evidence of clinical validity beyond the clustering.

free parameters (3)
  • number of contexts = 2
    Determined via clustering after PCA on health trajectory variables
  • MDP cost parameters
    Costs for different actions and states likely fitted or assigned based on data and assumptions
  • discount factor
    Standard in MDP but value not specified
assumptions (2)
  • domain assumption Markov property holds for patient state transitions
    Assumed in the CMDP model construction
  • ad hoc to paper The identified contexts represent stable subpopulations
    Based on clustering results
invented entities (1)
  • two risk contexts (lower- and higher-risk subpopulations)
    purpose: To stratify patients for context-specific policies
    Derived from data clustering without external validation mentioned

how reviews work

0 comments
Cite this review

Pith. "Pith review of Context-Aware Optimization of Follow-Up Intervals for Type 2 Diabetes Care Using Markov Decision Processes." pith.science (2026). https://pith.science/paper/YEO5YNPK

@misc{pith2026260619092,
  author       = {Pith},
  title        = {Pith review of: Context-Aware Optimization of Follow-Up Intervals for Type 2 Diabetes Care Using Markov Decision Processes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YEO5YNPK}},
  note         = {Machine review of arXiv:2606.19092}
}
read the original abstract

Chronic disease management relies on regular patient-provider interactions to follow-up on disease progression and control. For Type 2 Diabetes (T2D), current guidelines prescribe fixed time intervals between subsequent primary care visits for all patients, overlooking heterogeneity in clinical trajectories and patient characteristics. This study introduces a Contextual Markov Decision Process (CMDP) model to optimize subpopulation-specific follow-up interval decisions using Electronic Health Record (EHR) data from 22,154 T2D patients across 10 primary care clinics. Contexts are identified by: i) dimensionality reduction of variables representing the individual health trajectories utilizing Principal Component Analysis, and ii) assigning patients to contexts via principal components and additional patient-level features using clustering. Two distinct contexts emerged, representing a lower- and a higher-risk subpopulation. CMDP-derived policies recommend: (i) follow-up within 1 month if lab value at current visit is unmeasured; (ii) up to 3 months for elevated lab values or recent hospitalizations; and (iii) 6 to 12 months for sustained glycemic control, with shorter follow-up intervals for patients in high-risk context. The optimal policies achieved lower expected cumulative cost than benchmarks (e.g., in the higher-comorbidity context, the CMDP policy reduced cost by about 34.8%, and in the lower-comorbidity context by about 6.4%, relative to an American Diabetes Association-like fixed interval follow-up policy. These findings demonstrate how context-aware approaches can inform adaptive follow-up strategies, and have the potential to advance chronic care management in primary care by synthesizing machine learning and probabilistic decision models.

Figures

Figures reproduced from arXiv: 2606.19092 by the authors.

Figure 1
Figure 1. Overview of the methodological approach for deriving context-specific follow [PITH_FULL_IMAGE:figures/full_fig_p021_1.png] view at source ↗
Figure 2
Figure 2. Selection of principal components based on cumulative explained variance. [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Each row denotes a clinical state s=(glycemic status, short-horizon HbA1c [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: : Comparative policy evaluation by patient context. Expected cumulative cost [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: : Sensitivity analysis of cost parameters used in CMDP model. [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 1 canonical work pages

  1. [1]

    Diabetes care , volume=

    Introduction and methodology: standards of care in diabetes—2025 , author=. Diabetes care , volume=. 2025 , publisher=

    2025

  2. [2]

    Diabetes care , volume=

    Standards of care in diabetes—2023 , author=. Diabetes care , volume=

    2023

  3. [3]

    Diabetes, Obesity and Metabolism , volume=

    Effectiveness and cost-effectiveness of 3-monthly versus 6-monthly monitoring of well-controlled type 2 diabetes patients: a pragmatic randomised controlled patient-preference equivalence trial in primary care (EFFIMODI study) , author=. Diabetes, Obesity and Metabolism , volume=. 2014 , publisher=

    2014

  4. [4]

    BMC Endocrine Disorders , volume=

    Effectiveness of monthly and bimonthly follow-up of patients with well-controlled type 2 diabetes: a propensity score matched cohort study , author=. BMC Endocrine Disorders , volume=. 2019 , publisher=

    2019

  5. [5]

    Journal of diabetes and its complications , volume=

    Revisit frequency and its association with quality of care among diabetic patients: Translating Research Into Action for Diabetes (TRIAD) , author=. Journal of diabetes and its complications , volume=. 2014 , publisher=

    2014

  6. [6]

    Acta Diabetologica , volume=

    Analysis via Markov decision process to evaluate glycemic control strategies of a large retrospective cohort with type 2 diabetes: the ameliorate study , author=. Acta Diabetologica , volume=. 2020 , publisher=

    2020

  7. [7]

    Health Care Management Science , volume=

    Optimizing diabetes screening frequencies for at-risk groups , author=. Health Care Management Science , volume=. 2022 , publisher=

    2022

  8. [8]

    Management Science , volume=

    Personalized disease screening decisions considering a chronic condition , author=. Management Science , volume=. 2023 , publisher=

    2023

Show all 52 references
  1. [9]

    Scientific Reports , volume=

    Poor glycemic control and associated factors among patients with type 2 diabetes mellitus: a cross-sectional study , author=. Scientific Reports , volume=. 2023 , publisher=

    2023

  2. [10]

    Scientific reports , volume=

    Optimal treatment recommendations for diabetes patients using the Markov decision process along with the South Korean electronic health records , author=. Scientific reports , volume=. 2021 , publisher=

    2021

  3. [11]

    International Classification of Diseases, 11th Revision (ICD-11) , year =

  4. [12]

    Archives of internal medicine , volume=

    Encounter frequency and serum glucose level, blood pressure, and cholesterol level control in patients with diabetes mellitus , author=. Archives of internal medicine , volume=. 2011 , publisher=

    2011

  5. [13]

    Markov decision processes in practice , pages=

    Markov decision processes for screening and treatment of chronic diseases , author=. Markov decision processes in practice , pages=. 2017 , publisher=

    2017

  6. [14]

    Operations Research , volume=

    OR Forum—A POMDP approach to personalize mammography screening decisions , author=. Operations Research , volume=. 2012 , publisher=

    2012

  7. [15]

    Diabetes Research and Clinical Practice , pages=

    IDF global clinical practice recommendations for managing type 2 diabetes--2025 , author=. Diabetes Research and Clinical Practice , pages=. 2025 , publisher=

    2025

  8. [16]

    Type 2 Diabetes , year =

  9. [17]

    Glycemic Goals and Hypoglycemia: Standards of Care in Diabetes--2025 , journal =

    6. Glycemic Goals and Hypoglycemia: Standards of Care in Diabetes--2025 , journal =. 2025 , volume =. doi:10.2337/dc25-S006 , url =

    2025 doi

  10. [18]

    Journal of Xiangya Medicine , volume=

    Factors associated with poor glycemic control in patients with type 2 diabetes at the Bafoussam Regional Hospital: a preliminary cross-sectional study at the Bafoussam Regional Hospital (Cameroon) , author=. Journal of Xiangya Medicine , volume=. 2024 , publisher=

    2024

  11. [19]

    Endocrine Practice , volume=

    American Association of Clinical Endocrinology clinical practice guideline: developing a diabetes mellitus comprehensive care plan—2022 update , author=. Endocrine Practice , volume=. 2022 , publisher=

    2022

  12. [20]

    Glycemic targets: standards of care in diabetes—2023 , author=

    6. Glycemic targets: standards of care in diabetes—2023 , author=. Diabetes care , volume=. 2023 , publisher=

    2023

  13. [21]

    Clinical Diabetes , volume=

    Role of primary care clinicians in the management of patients with type 2 diabetes and cardiorenal diseases , author=. Clinical Diabetes , volume=. 2022 , publisher=

    2022

  14. [22]

    Introduction and application , author=

    Principal component analysis: a beginner's guide—I. Introduction and application , author=. Weather , volume=. 1990 , publisher=

    1990

  15. [23]

    Medical care , volume=

    A quantitative measure of continuity of care , author=. Medical care , volume=. 1977 , publisher=

    1977

  16. [24]

    Journal of General Internal Medicine , volume=

    Regularity and continuity of GP contacts and use of statins amongst people at risk of cardiovascular events , author=. Journal of General Internal Medicine , volume=. 2021 , publisher=

    2021

  17. [25]

    BMC family practice , volume=

    Effects of continuity of care on health outcomes among patients with diabetes mellitus and/or hypertension: a systematic review , author=. BMC family practice , volume=. 2021 , publisher=

    2021

  18. [26]

    Journal of Korean medical science , volume=

    Continuity of care for elderly patients with diabetes mellitus, hypertension, asthma, and chronic obstructive pulmonary disease in Korea , author=. Journal of Korean medical science , volume=

  19. [27]

    Therapeutics and clinical risk management , pages=

    Continuity of care with physicians and risk of subsequent hospitalization and end-stage renal disease in newly diagnosed type 2 diabetes mellitus patients , author=. Therapeutics and clinical risk management , pages=. 2018 , publisher=

    2018

  20. [28]

    BMC Health Services Research , volume=

    The cost-saving effect of continuity of care (COC): an analysis of institutional COC on diabetic treatment costs using panel 2SLS regressions , author=. BMC Health Services Research , volume=. 2021 , publisher=

    2021

  21. [29]

    BMC Health Services Research , volume=

    Greater temporal regularity of primary care visits was associated with reduced hospitalizations and mortality, even after controlling for continuity of care , author=. BMC Health Services Research , volume=. 2023 , publisher=

    2023

  22. [30]

    BMC Health Services Research , volume=

    Patient-level predictors of temporal regularity of primary care visits , author=. BMC Health Services Research , volume=. 2023 , publisher=

    2023

  23. [31]

    JAMA Network Open , volume=

    Primary care continuity, frequency, and regularity associated with Medicare savings , author=. JAMA Network Open , volume=. 2023 , publisher=

    2023

  24. [32]

    2019 , publisher=

    Hands-on machine learning with R , author=. 2019 , publisher=

    2019

  25. [33]

    Data mining and knowledge discovery , volume=

    Extensions to the k-means algorithm for clustering large data sets with categorical values , author=. Data mining and knowledge discovery , volume=. 1998 , publisher=

    1998

  26. [34]

    Journal of computational and applied mathematics , volume=

    Silhouettes: a graphical aid to the interpretation and validation of cluster analysis , author=. Journal of computational and applied mathematics , volume=. 1987 , publisher=

    1987

  27. [35]

    Biometrics , pages=

    A general coefficient of similarity and some of its properties , author=. Biometrics , pages=. 1971 , publisher=

    1971

  28. [36]

    Scientific reports , volume=

    Head-to-head comparison of clustering methods for heterogeneous data: a simulation-driven benchmark , author=. Scientific reports , volume=. 2021 , publisher=

    2021

  29. [37]

    Scientific Reports , volume=

    An investigation into epidemiological situations of COVID-19 with fuzzy K-means and K-prototype clustering methods , author=. Scientific Reports , volume=. 2023 , publisher=

    2023

  30. [38]

    Data & Knowledge Engineering , volume=

    A k-mean clustering algorithm for mixed numeric and categorical data , author=. Data & Knowledge Engineering , volume=. 2007 , publisher=

    2007

  31. [39]

    Public health , volume=

    Glycaemic control and associated factors among patients with diabetes at public health clinics in Johor, Malaysia , author=. Public health , volume=. 2016 , publisher=

    2016

  32. [40]

    Canadian Family Physician , volume=

    Relationship between GP visits and time spent in-hospital among insulin-dependent Canadians with type 2 diabetes , author=. Canadian Family Physician , volume=. 2020 , publisher=

    2020

  33. [41]

    Diabetes Care , volume=

    The association between poor glycemic control and health care costs in people with diabetes: a population-based study , author=. Diabetes Care , volume=. 2020 , publisher=

    2020

  34. [42]

    INQUIRY: The Journal of Health Care Organization, Provision, and Financing , volume=

    Prevalence of glycemic control and factors associated with poor glycemic control: A systematic review and meta-analysis , author=. INQUIRY: The Journal of Health Care Organization, Provision, and Financing , volume=. 2023 , publisher=

    2023

  35. [43]

    Endocrine Practice , volume=

    Consensus statement by the American Association of Clinical Endocrinologists and American College of Endocrinology on the comprehensive type 2 diabetes management algorithm--2020 executive summary , author=. Endocrine Practice , volume=. 2020 , publisher=

    2020

  36. [44]

    Diabetes Care in the Hospital: Standards of Care in Diabetes--2025 , journal =

    16. Diabetes Care in the Hospital: Standards of Care in Diabetes--2025 , journal =. 2025 , volume =

    2025

  37. [45]

    Diabetes care , volume=

    Diabetes and prediabetes and risk of hospitalization: the Atherosclerosis Risk in Communities (ARIC) Study , author=. Diabetes care , volume=. 2016 , publisher=

    2016

  38. [46]

    International journal of clinical practice , volume=

    Rates and risk of hospitalisation among patients with type 2 diabetes: retrospective cohort study using the UK General Practice Research Database linked to English Hospital Episode Statistics , author=. International journal of clinical practice , volume=. 2014 , publisher=

    2014

  39. [47]

    Diabetes & metabolism journal , volume=

    Glycemic variability: how do we measure it and why is it important? , author=. Diabetes & metabolism journal , volume=

  40. [48]

    Diabetes & metabolism , volume=

    The application of simple metrics in the assessment of glycaemic variability , author=. Diabetes & metabolism , volume=. 2018 , publisher=

    2018

  41. [49]

    Diabetes research and clinical practice , volume=

    Long-term glycemic variability and risk of adverse health outcomes in patients with diabetes: A systematic review and meta-analysis of cohort studies , author=. Diabetes research and clinical practice , volume=. 2022 , publisher=

    2022

  42. [50]

    Cardiovascular diabetology , volume=

    Glycemic variability: adverse clinical outcomes and how to improve it? , author=. Cardiovascular diabetology , volume=. 2020 , publisher=

    2020

  43. [51]

    2009 , publisher=

    The elements of statistical learning , author=. 2009 , publisher=

    2009

  44. [52]

    Diabetologia , volume=

    Impact of age, age at diagnosis and duration of diabetes on the risk of macrovascular and microvascular complications and death in type 2 diabetes , author=. Diabetologia , volume=. 2014 , publisher=

    2014

Pith tools

Reviewed June 26, 2026 · model on record in the stance chip above.