Using Machine Learning to Individualize Treatment Effect Estimation: Challenges and Opportunities

The use of data from randomized clinical trials to justify treatment decisions for real-world patients is the current state of the art. It relies on the assumption that average treatment effects from the trial can be extrapolated to patients with personal and/or disease characteristics different from those treated in the trial. Yet, because of heterogeneity of treatment effects between patients and between the trial population and real-world patients, this assumption may not be correct for many patients. Using machine learning to estimate the expected conditional average treatment effect (CATE) in individual patients from observational data offers the potential for more accurate estimation of the expected treatment effects in each patient based on their observed characteristics. In this review, we discuss some of the challenges and opportunities for machine learning to estimate CATE, including ensuring identification assumptions are met, managing covariate shift, and learning without access to the true label of interest. We also discuss the potential applications as well as future work and collaborations needed to further improve identification and utilization of CATE estimates to increase patient benefit.

Can machine learning augment clinician adjudication of events in cardiovascular trials? A case study of major adverse cardiovascular events (MACE) across CVRM trials

Background and introduction

Accurate identification of clinical outcome events is critical to obtaining reliable results in cardiovascular outcomes trials (CVOTs). Current processes for event adjudication are expensive and hampered by delays. As part of a larger project to more reliably identify outcomes, we evaluated the use of machine learning to automate event adjudication using data from the SOCRATES trial (NCT01994720), a large randomized trial comparing ticagrelor and aspirin in reducing risk of major cardiovascular events after acute ischemic stroke or transient ischemic attack (TIA).

Purpose

We studied whether machine learning algorithms could replicate the outcome of the expert adjudication process for clinical events of ischemic stroke and TIA. Could classification models be trained on historical CVOT data and demonstrate performance comparable to human adjudicators?

Methods

Using data from the SOCRATES trial, multiple machine learning algorithms were tested using grid search and cross validation. Models tested included Support Vector Machines, Random Forest and XGBoost. Performance was assessed on a validation subset of the adjudication data not used for training or testing in model development. Metrics used to evaluate model performance were Receiver Operating Characteristic (ROC), Matthews Correlation Coefficient, Precision and Recall. The contribution of features, attributes of data used by the algorithm as it is trained to classify an event, that contributed to a classification were examined using both Mutual Information and Recursive Feature Elimination.

Results

Classification models were trained on historical CVOT data using adjudicator consensus decision as the ground truth. Best performance was observed on models trained to classify ischemic stroke (ROC 0.95) and TIA (ROC 0.97). Top ranked features that contributed to classification of Ischemic Stroke or TIA corresponded to site investigator decision or variables used to define the event in the trial charter, such as duration of symptoms. Model performance was comparable across the different machine learning algorithms tested with XGBoost demonstrating the best ROC on the validation set for correctly classifying both stroke and TIA.

Conclusions

Our results indicate that machine learning may augment or even replace clinician adjudication in clinical trials, with potential to gain efficiencies, speed up clinical development, and retain reliability. Our current models demonstrate good performance at binary classification of ischemic stroke and TIA within a single CVOT with high consistency and accuracy between automated and clinician adjudication. Further work will focus on harmonizing features between multiple historical clinical trials and training models to classify several different endpoint events across trials. Our aim is to utilize these clinical trial datasets to optimize the delivery of CVOTs in further cardiovascular drug development.

Funding Acknowledgement

Type of funding sources: Private company. Main funding source(s): AstraZenca Plc

The relative efficiency of schizophrenia health care systems: an international comparison using data envelopment analysis

AIM

To compare the health economic efficiency of health care systems across nations, within the area of schizophrenia, using a data envelopment analysis (DEA) approach.

METHODS

The DEA was performed using countries as decision-making units, schizophrenia disease investment (cost of disease as a percentage of total health care expenditure) as the input, and disability-adjusted life years (DALYs) per patient due to schizophrenia as the output. Data were obtained from the Global Burden of Disease 2017 study, the World Bank Group, and a literature search of the PubMed database.

RESULTS

Data were obtained for 44 countries; of these, 34 had complete data and were included in the DEA. Disease investment (percentage of total health care expenditure) ranged from 1.11 in Switzerland to 6.73 in Thailand. DALYs per patient ranged from 0.621 in Lithuania to 0.651 in Malaysia. According to the DEA, countries with the most efficient schizophrenia health care were Lithuania, Norway, Switzerland and the US (all with efficiency score 1.000). The least efficient countries were Malaysia (0.955), China (0.959) and Thailand (0.965).

LIMITATIONS

DEA findings depend on the countries and variables that are included in the dataset.

CONCLUSIONS

In this international DEA, despite the difference in schizophrenia disease investment across countries, there was little difference in output as measured by DALYs per patient. Potentially, Lithuania, Norway, Switzerland and the US should be considered 'benchmark' countries by policy makers, thereby providing useful information to countries with less efficient systems.

Machine learning for clinical trials in the era of COVID-19

The world is in the midst of a pandemic. We still know little about the disease COVID-19 or about the virus (SARS-CoV-2) that causes it. We do not have a vaccine or a treatment (aside from managing symptoms). We do not know if recovery from COVID-19 produces immunity, and if so for how long, hence we do not know if "herd immunity" will eventually reduce the risk or if a successful vaccine can be developed - and this knowledge may be a long time coming. In the meantime, the COVID-19 pandemic is presenting enormous challenges to medical research, and to clinical trials in particular. This paper identifies some of those challenges and suggests ways in which machine learning can help in response to those challenges. We identify three areas of challenge: ongoing clinical trials for non-COVID-19 drugs; clinical trials for repurposing drugs to treat COVID-19, and clinical trials for new drugs to treat COVID-19. Within each of these areas, we identify aspects for which we believe machine learning can provide invaluable assistance.

The Tendril Plot-a novel visual summary of the incidence, significance and temporal aspects of adverse events in clinical trials

Background

In contrast to efficacy, safety hypotheses of clinical trials are not always pre-specified, and therefore, the safety interpretation work of a trial tends to be more exploratory, often reactive, and the analysis more statistically and graphically challenging.

Methods

We introduce a new means of visualizing the adverse event data across an entire clinical trial.

Results

The approach overcomes some of the current limitations of adverse event analysis and streamlines the way safety data can be explored, interpreted and analyzed. Using a phase II study, we describe and exemplify how the tendril plot effectively summarizes the time-resolved safety profile of two treatment arms in a single plot and how that can provide scientists with a trial safety overview that can support medical decision making.

Conclusion

To our knowledge, the tendril plot is the only way to graphically show important treatment differences with preserved temporal information, across an entire clinical trial, in a single view.

A systematic literature review identifying associations between outcomes and quality of life (QoL) or healthcare resource utilization (HCRU) in schizophrenia

AIMS

There have been no systematic literature reviews (SLRs) evaluating the identified association between outcomes (e.g. clinical, functional, adherence, societal burden) and Quality-of-Life (QoL) or Healthcare Resource Utilization (HCRU) in schizophrenia. The objective of this study was to conduct a SLR of published data on the relationship between outcomes and QoL or HCRU.

MATERIALS AND METHODS

Electronic searches were conducted in Embase and Medline, for articles which reported on the association between outcomes and QoL or HCRU. Inclusion and exclusion criteria were applied to identify the most relevant articles and studies and extract their data. A summary table was developed to illustrate the strength of associations, based on p-values and correlations.

RESULTS

One thousand and two abstracts were retrieved; five duplicates were excluded; 997 abstracts were screened and 95 references were retained for full-text screening. Thrirty-one references were included in the review. The most commonly used questionnaire, which also demonstrated the strongest associations (defined as a p < 0.0001 and/or correlation ±0.70), was the Positive and Negative Syndrome Scale (PANSS) associated with HCRU and QoL (the SF-36, the Schizophrenia Quality-of-Life questionnaire [S-QOL-18], the Quality-of-Life Scale [QLS]). Other robust correlations included the Clinical Global Impression-Severity (CGI-S) with QoL (EQ5D), relapse with HCRU, and remission with QoL (EQ5D). Lastly, functioning (Work Rehabilitation Questionnaire [WORQ] and Personal and Social Performance Scale [PSP]) was found to be associated to QoL (QLS and Subjective Well-being under Neuroleptics Questionnaire [SWN]).

LIMITATIONS

This study included data from an 11-year period, and other instruments less frequently used may be further investigated.

CONCLUSIONS

The evidence suggests that the PANSS is the clinical outcome that currently provides the most frequent and systematic associations with HCRU and QoL endpoints in schizophrenia.

Distinguishing prognostic and predictive biomarkers: an information theoretic approach

Motivation

The identification of biomarkers to support decision-making is central to personalized medicine, in both clinical and research scenarios. The challenge can be seen in two halves: identifying predictive markers, which guide the development/use of tailored therapies; and identifying prognostic markers, which guide other aspects of care and clinical trial planning, i.e. prognostic markers can be considered as covariates for stratification. Mistakenly assuming a biomarker to be predictive, when it is in fact largely prognostic (and vice-versa) is highly undesirable, and can result in financial, ethical and personal consequences. We present a framework for data-driven ranking of biomarkers on their prognostic/predictive strength, using a novel information theoretic method. This approach provides a natural algebra to discuss and quantify the individual predictive and prognostic strength, in a self-consistent mathematical framework.

Results

Our contribution is a novel procedure, INFO+, which naturally distinguishes the prognostic versus predictive role of each biomarker and handles higher order interactions. In a comprehensive empirical evaluation INFO+ outperforms more complex methods, most notably when noise factors dominate, and biomarkers are likely to be falsely identified as predictive, when in fact they are just strongly prognostic. Furthermore, we show that our methods can be 1-3 orders of magnitude faster than competitors, making it useful for biomarker discovery in 'big data' scenarios. Finally, we apply our methods to identify predictive biomarkers on two real clinical trials, and introduce a new graphical representation that provides greater insight into the prognostic and predictive strength of each biomarker.

Availability and implementation

R implementations of the suggested methods are available at https://github.com/sechidis.

Supplementary information

Supplementary data are available at Bioinformatics online.

Sleep Tracking and Exercise in Patients With Type 2 Diabetes Mellitus (Step-D): Pilot Study to Determine Correlations Between Fitbit Data and Patient-Reported Outcomes

Background

Few studies assessing the correlation between patient-reported outcomes and patient-generated health data from wearable devices exist.

Objective

The aim of this study was to determine the direction and magnitude of associations between patient-generated health data (from the Fitbit Charge HR) and patient-reported outcomes for sleep patterns and physical activity in patients with type 2 diabetes mellitus (T2DM).

Methods

This was a pilot study conducted with adults diagnosed with T2DM (n=86). All participants wore a Fitbit Charge HR for 14 consecutive days and completed internet-based surveys at 3 time points: day 1, day 7, and day 14. Patient-generated health data included minutes asleep and number of steps taken. Questionnaires assessed the number of days of exercise and nights of sleep problems per week. Means and SDs were calculated for all data, and Pearson correlations were used to examine associations between patient-reported outcomes and patient-generated health data. All respondents provided informed consent before participating.

Results

The participants were predominantly middle-aged (mean 54.3, SD 13.3 years), white (80/86, 93%), and female (50/86, 58%). Use of oral T2DM medication correlated with the number of mean steps taken (r=.35, P=.001), whereas being unaware of the glycated hemoglobin level correlated with the number of minutes asleep (r=−.24, P=.04). On the basis of the Fitbit data, participants walked an average of 4955 steps and slept 6.7 hours per day. They self-reported an average of 2.0 days of exercise and 2.3 nights of sleep problems per week. The association between the number of days exercised and steps walked was strong (r=.60, P<.001), whereas the association between the number of troubled sleep nights and minutes asleep was weaker (r=.28, P=.02).

Conclusions

Fitbit and patient-reported data were positively associated for physical activity as well as sleep, with the former more strongly correlated than the latter. As extensive patient monitoring can guide clinical decisions regarding T2DM therapy, passive, objective data collection through wearables could potentially enhance patient care, resulting in better patient-reported outcomes.

When insulin degludec enhances quality of life in patients with type 2 diabetes: a qualitative investigation

Background

Anecdotal reports suggest that insulin degludec (IDeg) may offer unique health-related quality of life (HRQoL) benefits. As the nature of these benefits remain unclear, this study utilized qualitative research methods to investigate and elucidate the experience of “feeling better” after initiating IDeg.

Methods

Twenty adults with type 2 diabetes (T2D) who reported “feeling better” on IDeg for > 3 months participated in 90-min interviews. One focus group and nine telephone interviews were conducted at two sites in the United States (US) and one focus group was conducted in Switzerland. Patients were ≥ 18 years of age, did not take mealtime insulin, and had switched to IDeg from another basal insulin. Discussions were audio-recorded, transcribed and translated (Swiss German). Utilizing grounded theory, transcripts were analyzed by sorting quotes into concepts using thematic analysis.

Results

Participants' mean age was 66 years and the average duration of T2D was 17.6 years. Mean duration of IDeg use was 1.45 years. Four major factors were identified as key contributors to patients’ sense of “feeling better”: 1) reduced sense of diabetes as burdensome and requiring excessive attention; 2) enhanced feelings of adaptability and freedom; 3) heightened sense of security, especially regarding concerns about hypoglycemia; and 4) greater sense of physical well-being (greater energy/less fatigue). Content saturation was achieved. Generally, patients from the US sites were more focused on medical results than Swiss patients, who were more likely to identify IDeg’s effect on overall HRQoL. A limitation of the study was that the population was primarily white, > 60 and otherwise healthy (no comorbid physical or mental condition).

Conclusions

A group of patients with T2D, who had switched to IDeg from another basal insulin, reported HRQoL benefits which were attributed to both diabetes-specific improvements (feeling less burdened by day-to-day diabetes demands) and non-specific gains (greater energy). The conclusions may have limited transferability due to the characteristics of the sample population and further research is needed.

Insulin degludec versus insulin glargine U100 for patients with type 1 or type 2 diabetes in the US: a budget impact analysis with rebate tables

BACKGROUND AND AIMS

Drug rebates are almost universally negotiated privately between the manufacturer and the payer in the US. The aim of the present study was to illustrate the use of a "rebate table" to improve the transparency and utility of published budget impact analyses in the US by modeling ranges of hypothetical rebates for two comparators. Worked examples were conducted to illustrate the budgetary implications of using insulin degludec (IDeg) relative to insulin glargine (IGlar) U100 in patients with type 1 or 2 diabetes.

METHODS

A short-term (1-year) budget impact model was developed to evaluate the costs of switching to IDeg from IGlar in patients with type 1 or 2 diabetes on basal-bolus and basal-only insulin, respectively. The analysis used insulin dose and hypoglycemia data from recent randomized trials, data on the prevalence of diabetes, and estimates of the proportion of patients using each insulin regimen. The model was configured to run multiple rebate scenarios to generate a rebate table in a hypothetical 1 million member commercial plan.

RESULTS

Relative to IGlar, IDeg resulted in reductions in non-severe and severe hypoglycemia incidence and costs both in patients with type 1 and patients with type 2 diabetes. Insulin acquisition costs were higher, and respective rebates of 7.3% and 10.6% were required for IDeg to break-even with IGlar at the full list price. Incremental per member per month IDeg costs without a rebate were USD 0.04 in type 1 diabetes and USD 0.80 in type 2 diabetes.

CONCLUSIONS

Using IDeg instead of IGlar at list price could result in a modest increase in costs when considering insulin and hypoglycemia costs alone, but modest incremental rebates with IDeg would result in cost neutrality relative to IGlar. In addition, IDeg would result in reduced incidence of severe and non-severe hypoglycemia.

The burden of severe hypoglycemia in type 2 diabetes

AIMS

More than 29 million people in the US have type 2 diabetes mellitus (T2DM), a chronic metabolic disorder characterized by a progressive deterioration of glucose control, which eventually requires insulin. Abnormally low levels of blood glucose, a feared side-effect of insulin treatment, may cause severe hypoglycemia (SHO), leading to emergency department (ED) admission, hospitalization, and long-term complications; these, in turn, drive up the costs of T2DM. This study's objective was to estimate the prevalence and costs of SHO-related hospitalizations and their additional longer-term impacts on patients with T2DM using insulin.

METHODS

Using Truven MarketScan claims, we identified adult T2DM patients using basal and basal-bolus insulin regimens who were hospitalized for SHO (inpatient SHO patients) during 2010-2015. Two comparison groups were defined: those with outpatient SHO-related encounters only, including ED visits without hospitalization (outpatient SHO patients), and those with no SHO- or acute hyperglycemia-related events (comparison patients). Lengths of stay and SHO-related hospitalization costs were estimated, and propensity score and inverse probability weighting methods were used to adjust for baseline differences across the groups to evaluate longer-term impacts.

RESULTS

We identified 66,179 patients using basal and 81,876 patients using basal-bolus insulin, of which ∼1.1% (basal) to 3.2% (basal-bolus) experienced at least one SHO-related hospitalization. Among those who experienced SHO (i.e. those in the inpatient and outpatient SHO groups), 27% (basal) and 40% (basal-bolus) experienced at least one SHO-related hospitalization. One-third of basal and about one-quarter of basal-bolus patients were admitted directly to the hospital; the remainder were first assessed or treated in the ED. Inpatient SHO patients using basal insulin stayed in the hospital, including time in the ED, for 2.8 days and incurred $6896 in costs; patients using basal-bolus insulin stayed in the hospital for 2.6 days and incurred costs of $5802. Forty-to-fifty percent of inpatient SHO patients were hospitalized again for SHO. Inpatient SHO patients using basal insulin incurred significantly higher monthly costs after their initial SHO-related hospitalization than patients in the other two groups ($2935 vs $1819 and $1638), corresponding to 61% and 79% higher monthly costs; patients using basal-bolus insulin also incurred significantly higher monthly costs than patients in the other groups ($3606 vs $2731 and $2607), corresponding to 32% and 38% higher monthly costs.

LIMITATIONS

These analyses excluded patients who did not seek ED or hospital care when faced with SHO; events may have been miscoded; and we were not able to account for clinical characteristics associated with SHO, such as insulin dose and duration of diabetes, or unmeasured confounders.

CONCLUSIONS

The burden associated with SHO is not negligible. Nearly one in three patients using only basal insulin and one in four patients using basal-bolus regimens who experienced SHO were hospitalized at least once due to SHO. Not only did those patients incur the costs of their SHO hospitalization, but they also incurred at least $1,116 (62%) and $875 (70%) more per month than outpatient SHO or comparison patients. Reducing SHO events can help decrease the burden associated with SHO among patients with T2DM.

The burden of severe hypoglycemia in type 1 diabetes

AIMS

Approximately 1.25 million people in the US have type 1 diabetes mellitus (T1DM), a chronic metabolic disease that develops from the body's inability to produce insulin, and requires life-long insulin therapy. Poor insulin adherence may cause severe hypoglycemia (SHO), leading to hospitalization and long-term complications; these, in turn, drive up costs of SHO and T1DM overall. This study's objective was to estimate the prevalence and costs of SHO-related hospitalizations and their additional longer-term impacts on patients with T1DM using basal-bolus insulin.

METHODS

Using Truven MarketScan claims, we identified adult T1DM patients using basal-bolus insulin regimens who were hospitalized for SHO (inpatient SHO patients) during 2010-2015. Two comparison groups were defined: those with outpatient SHO-related encounters only, including emergency department (ED) visits without hospitalization (outpatient SHO patients), and those with no SHO- or acute hyperglycemia-related events (comparison patients). Lengths of stay and SHO-related hospitalization costs were estimated and propensity score and inverse probability weighting methods were used to adjust for baseline differences across the groups to evaluate longer-term impacts.

RESULTS

We identified 8,734 patients, of which 4.2% experienced at least one SHO-related hospitalization. Among those who experienced SHO (i.e. of those in the inpatient and outpatient SHO groups), 31% experienced at least one SHO-related hospitalization, while 9% were treated in the ED without subsequent hospitalization. Approximately 79% of patients were admitted directly to the hospital; the remainder were first assessed or treated in the ED. The inpatient SHO patients stayed in the hospital, including time in the ED, for 1.7 days and incurred $3551 in costs. About one-third of patients were hospitalized again for SHO. Inpatient SHO patients incurred significantly higher monthly costs after their initial SHO-related hospitalization than patients in the two other groups ($2084 vs $1313 and $1372), corresponding to 59% or 52% higher monthly costs for inpatient SHO patients.

LIMITATIONS

These analyses excluded patients who did not seek ED or hospital care when faced with SHO; events may have been miscoded; and we were not able to account for clinical characteristics associated with SHO, such as insulin dose and duration of diabetes, or unmeasured confounders.

CONCLUSIONS

The burden associated with SHO is not negligible. About 4% of T1DM patients using basal-bolus insulin regimens are hospitalized at least once due to SHO. Not only did those patients incur the costs of their SHO hospitalization, but they also incur red at least $712 (52%) more in costs per month after their hospitalization than outpatient SHO or comparison patients. Reducing SHO events can help decrease the burden associated with SHO among patients with T1DM.

Budget impact of treating commercially insured type 1 and type 2 diabetes patients in the United States with insulin degludec compared to insulin glargine

OBJECTIVE

To quantify the annual budget impact if all US commercially insured type 1 diabetes mellitus patients on basal-bolus therapy (T1DM_BBT), type 2 diabetes mellitus patients on basal-oral therapy (T2DM_BOT), and type 2 diabetes mellitus patients on basal-bolus therapy (T2DM_BBT) switched from insulin glargine (IGlar) to insulin degludec (IDeg).

METHODS

A short-term (1 year) budget impact model was developed to evaluate the costs of IDeg vs. IGlar in three treatment groups (T1DM_BBT, insulin-naïve T2DM_BOT, and T2DM_BBT) through a simulation for a potential US health plan population of 35 million. The analysis captured direct medical costs associated with insulin treatment (insulin, needles, and self-monitored glucose testing) and costs related to managing hypoglycemic episodes. There were a total of 59,780 T1DM_BBT patients, 383,145 T2DM_BOT patients, and 171,325 T2DM_BBT patients expected to be using long-acting insulin. A sensitivity analysis on the entire US population was also conducted.

RESULTS

Among T1DM_BBT patients, IDeg was associated with an annual cost savings of -$357.13 per patient per year (PPPY), driven primarily by reduced insulin utilization. IDeg was also found to be cost saving among T2DM_BOT patients (-$1206.61 PPPY), driven primarily by reductions in the cost of treating severe hypoglycemic episodes. Among T2DM_BBT patients, IDeg was associated with an additional cost to the plan of $1420.04 PPPY; however, this result was driven by a higher insulin dose for IDeg compared to IGlar. Overall, IDeg demonstrated cost savings of $240 million per year, which accounted for total cost savings of 3.5% vs. IGlar.

CONCLUSIONS

The results of this analysis suggest that the reduced insulin utilization and fewer hypoglycemic episodes associated with IDeg may translate into reduced costs for payers. The model is limited by simplification of a complex disease state and assumptions surrounding disease state, treatment patterns, and costs. Therefore, results may not accurately reflect actual health plans or real-world practice patterns.

Disentangling Prognostic and Predictive Biomarkers Through Mutual Information

We study information theoretic methods for ranking biomarkers. In clinical trials, there are two, closely related, types of biomarkers: predictive and prognostic, and disentangling them is a key challenge. Our first step is to phrase biomarker ranking in terms of optimizing an information theoretic quantity. This formalization of the problem will enable us to derive rankings of predictive/prognostic biomarkers, by estimating different, high dimensional, conditional mutual information terms. To estimate these terms, we suggest efficient low dimensional approximations. Finally, we introduce a new visualisation tool that captures the prognostic and the predictive strength of a set of biomarkers. We believe this representation will prove to be a powerful tool in biomarker discovery.

It's a long shot, but it just might work! Perspectives on the future of medicine

What does the future of medicine hold? We asked six researchers to share their most ambitious and optimistic views of the future, grounded in the present but looking out a decade or more from now to consider what's possible. They paint a picture of a connected and data-driven world in which patient value, patient feedback, and patient empowerment shape a continually learning system that ensures each patient's experience contributes to the improved outcome of every patient like them, whether it be through clinical trials, data from consumer devices, hacking their medical devices, or defining value in thoughtful new ways.

Understanding Post-Prandial Hyperglycemia in Patients with Type 1 and Type 2 Diabetes: A Web-based Survey in Germany, the UK, and USA

INTRODUCTION

To explore how patients with diabetes experience post-prandial hyperglycemia (PPH) or elevated blood glucose (BG) following a meal.

METHODS

A web-based survey of patients with type 1 or type 2 diabetes using bolus insulin in Germany, the USA, and the UK was conducted.

RESULTS

A total of 906 respondents completed the survey. PPH was a frequent occurrence among patients with type 1 and type 2 diabetes; 61.9% of respondents had experienced PPH in the past week, and differences by diabetes type were not significant. More than half of the respondents reported that they knew they were experiencing PPH because they had measured their BG (64.8%) and/or because they "just didn't feel right" (51.9%). The most frequently reported reasons given for PPH were eating more fat/sugar than estimated (31.2%) and over-eating in terms of their calculated bolus insulin dose (30.4%). The most common situations/factors contributing to PPH were stress (27.4%), eating at a restaurant (24.9%), being busy (21.1%), and/or feeling tired (19.2%). The most frequent corrective actions respondents took following PPH were testing BG and taking bolus insulin based on the reading (62.0%), and/or eating less/more carefully at their next meal or snack (18.8%). Additionally, significant differences in the reasons and contributing factors given for PPH and corrective actions following PPH, as well as emotions experienced when taking bolus insulin, were found by diabetes type.

CONCLUSION

These findings shed light on how patients with diabetes experience and manage PPH on a day-to-day basis and have implications for improving diabetes self-management. Clinicians and diabetes educators should help patients address eating habits and lifestyle issues that may contribute to PPH.

FUNDING

This study was sponsored by Novo Nordisk.

The Economic Burden of Post-prandial Hyperglycemia (PPH) Among People with Type 1 and Type 2 Diabetes in Three Countries

INTRODUCTION

Post-prandial hyperglycemia (PPH) among people with diabetes is a well-known clinical challenge to diabetes management. While the economic burden of diabetes is well studied, little is known about economic costs specific to PPH. The purpose of this study was to investigate costs of PPH related to work, diabetes management, and use of healthcare resources among people with diabetes taking bolus insulin.

METHODS

Data were collected in a web survey of 906 adults with type 1 (39%) and type 2 (61%) diabetes taking bolus insulin in Germany (34%), the UK (26%), and the USA (40%).

RESULTS

Sixty-two percent of respondents experienced PPH in the past week, and respondents averaged 1.7 episodes per week. Working respondents indicated that PPH affected their work productivity: 27% missed work time and 71% experienced work productivity issues while at work due to a recent episode of PPH. In terms of diabetes management, respondents with PPH in the past week measured their blood glucose (BG) more frequently than those without PPH (3.7 vs. 2.5 times/day, P < 0.001). PPH was also significantly associated with greater use of healthcare resources. Compared to those without PPH, respondents with PPH reported greater contact with healthcare professionals related to diabetes in the past year (5.5 vs. 4.4 visits, P < 0.001; 2.7 vs. 1.4 calls/emails, P < 0.001) and were more likely to report medical complications related to diabetes (72% vs. 55%, P < 0.001). Average annual costs associated with PPH due to missed work time, additional BG test strips, and physician visits were estimated to be $1239 USD per employed person in the USA.

CONCLUSION

Results indicate that PPH is associated with greater economic costs and that reducing the incidence of PPH would help mitigate such costs. Additional research is needed to better understand costs associated with PPH that may be more difficult to measure, as well as more long-term impacts and costs.

FUNDING

Novo Nordisk.

Taming EHR data: using semantic similarity to reduce dimensionality

Medical care data is a valuable resource that can be used for many purposes including managing and planning for future health needs as well as clinical research. However, the heterogeneity and complexity of medical data can be an obstacle in applying data mining techniques. Much of the potential value of this data therefore goes untapped. In this paper we have developed a methodology that reduces the dimensionality of primary care data, in order to make it more amenable to visualisation, mining and clustering. The methodology involves employing a combination of ontology-based semantic similarity and principal component analysis (PCA) to map the data into an appropriate and informative low dimensional space. Throughout the study, we had access to anonymised patient data from primary care in Salford, UK. The results of our application of this methodology show that diagnosis codes in primary care data can be used to map patients into an informative low dimensional space, which in turn provides the opportunity to support further data exploration and medical hypothesis formulation.

Quantifying the longitudinal value of healthcare record collections for pharmacoepidemiology

We introduce an information score for longitudinal healthcare record data, specifically in the monitoring of chronic conditions. The score is designed to capture the value of different observation patterns in terms of shaping and testing clinical epidemiological hypotheses. The score is first developed for the simple case where equally spaced observations are most informative, then extended to a more context-specific version where the optimal density of observations can be elicited. It can be interpreted as a measure of the average quantity of information provided by each observation in an individual's time course, where information is lost whenever the observation density deviates from a defined optimal density. We illustrate the score on routine healthcare records from the population of Salford, UK - focusing on repeat testing of liver function in people with common long-term conditions. We demonstrate validity of the score in terms of concordance between score levels and clinically meaningful patterns of repeat testing.

Secondary prophylaxis with recombinant activated factor VII improves health-related quality of life of haemophilia patients with inhibitors

Haemophilia patients with inhibitors characteristically have impaired joint function and reduced health-related quality of life (HRQoL). This analysis examined whether secondary prophylaxis with recombinant activated factor VII (rFVIIa) improves HRQoL vs. conventional on-demand therapy in patients with haemophilia with inhibitors and frequent bleeds. After a 3-month preprophylaxis period, 22 patients received daily rFVIIa prophylaxis (90 or 270 microg kg(-1)) for 3 months, followed by 3 months' postprophylaxis. Days of hospitalization, absence from school/work and mobility aids requirements were recorded. HRQoL was assessed by EuroQoL (EQ-5D) questionnaire, visual analogue scale (VAS), derived Time to Trade-Off (TTO) scores and Quality Adjusted Life Years (QALYs). rFVIIa prophylaxis significantly (P < 0.0001) reduced bleeding frequency vs. prior on-demand therapy. Hospitalization (5.9% vs. 13.5%; P = 0.0026) and absenteeism from school/work (16.7% vs. 38.7%; P = 0.0127) decreased during prophylaxis; these effects tended to be maintained during postprophylaxis. HRQoL (evaluated by EQ-5D) tended to improve during and after rFVIIa prophylaxis. Notably, pain decreased and mobility increased in 40.9% and 27.3% of patients, respectively, at the end of the postprophylaxis period vs. preprophylaxis. Median VAS score increased from 66 to 73 (P = 0.048), and TTO scores suggested better HRQoL (0.62 vs. 0.76; P = 0.054) during postprophylaxis than preprophylaxis. Small to moderate changes in effect sizes were reported for VAS and TTO scores. Median QALYs were 0.68 (VAS) and 0.73 (TTO). Reductions in bleeding frequency with secondary rFVIIa prophylaxis were associated with improved HRQoL vs. on-demand therapy.