Generating real-world data from health records: design of a patient-centric study in multiple sclerosis using a commercial health records platform

Objective

The FlywheelMS study will explore the use of a real-world health record data set generated by PicnicHealth, a patient-centric health records platform, to improve understanding of disease course and patterns of care for patients with multiple sclerosis (MS).

Materials and Methods

The FlywheelMS study aims to enroll 5000 adults with MS in the United States to create a large, deidentified, longitudinal data set for clinical research. PicnicHealth obtains health records, including paper charts, electronic health records, and radiology imaging files from any healthcare site. Using a large-scale health record processing pipeline, PicnicHealth abstracts standard and condition-specific data elements from structured (eg, laboratory test results) and unstructured (eg, narrative) text and maps these to standardized medical vocabularies. Researchers can use the resulting data set to answer empirical questions and study participants can access and share their harmonized health records using PicnicHealth’s web application.

Results

As of November 24, 2020, more than 4176 participants from 49 of 50 US states have enrolled in the FlywheelMS study. A median of 200 pages of records have been collected from 14 different doctors over 8 years per participant. Abstraction precision, established through inter-abstractor agreement, is as high as 97.8% when identifying and mapping data elements to a standard ontology.

Conclusion

Using a commercial health records platform, the FlywheelMS study is generating a real-world, multimodal data set that could provide valuable insights about patients with MS. This approach to data collection and abstraction is disease-agnostic and could be used to address other clinical research questions in the future.

Huntington's Disease in Israel: A Population-Based Study Using 20 Years of Routinely-Collected Healthcare Data

BACKGROUND

Huntington's disease (HD) is a rare, genetic, neurodegenerative disease. Obtaining population-level data on epidemiology and disease management is challenging.

OBJECTIVE

To investigate the epidemiology, clinical manifestations, treatment, and healthcare utilization of patients with HD in Israel.

METHODS

Retrospective population-based cohort study, including 20 years of routinely collected data from Maccabi Healthcare Services, an insurer and healthcare provider for one-quarter of the Israeli population.

RESULTS

The study cohort included 109 adult patients (aged ≥18 years) diagnosed with HD, with mean age of 49.9 years and 56%females. The most common HD-related conditions were anxiety (40%), behavioral problems (34%), sleep disorders (21%), and falls (13%). Annual incidence rates for HD ranged from 0.17 to 1.34 per 100,000 from 2000 to 2018; the 2018 crude prevalence in adults was 4.36 per 100,000. Median survival from diagnosis was approximately 12 years (95%CI: 10.4-15.3). The most frequent symptomatic treatments were antidepressants (69%), antipsychotics (63%), and tetrabenazine (63%), the only drug approved for the treatment of HD chorea in Israel during the examined period. Patterns of healthcare utilization changed as disease duration increased, reflected by increased frequency of emergency department visits and home visits.

CONCLUSION

This retrospective population-based study provides insights into the prevalence, incidence, clinical profile, survival, and resource utilization of patients with HD in ethnically diverse Israel. The findings in this study are generally consistent with the international literature and demonstrate the value of routinely collected healthcare data as a complementary resource in HD research.

Comparative effectiveness analysis between entrectinib clinical trial and crizotinib real-world data in ROS1+ NSCLC

Aim: Generating direct comparative evidence in prospective randomized trials is difficult for rare diseases. Real-world cohorts may supplement control populations. Methods: Entrectinib-treated adults with advanced ROS1 fusion-positive NSCLC (n = 94) from Phase I/II trials (ALKA-372-001 [EudraCT2012-00148-88], STARTRK-1 [NCT02097810], and STARTRK-2 [NCT02568267]) were compared with a real-world crizotinib-treated cohort (n = 65). Primary end point, time-to-treatment discontinuation (TTD); secondary end points, PFS and OS. Results: Median (95% CI) weighted TTD: 12.9 (9.9-17.4) months for entrectinib; 8.8 (6.2-9.9) months for crizotinib (weighted hazard ratio: 0.72 [0.51-1.02]). Median OS with entrectinib was not reached, weighted median OS with crizotinib was 18.5 (15.1-19.9) months. Conclusion: Entrectinib administered in clinical trials may be associated with longer TTD than a real-world crizotinib population.

Use of Patient Health Records to Quantify Drug-Related Pro-arrhythmic Risk

There is an increasing expectation that computational approaches may supplement existing human decision-making. Frontloading of models for cardiac safety prediction is no exception to this trend, and ongoing regulatory initiatives propose use of high-throughput in vitro data combined with computational models for calculating proarrhythmic risk. Evaluation of these models requires robust assessment of the outcomes. Using FDA Adverse Event Reporting System reports and electronic healthcare claims data from the Truven-MarketScan US claims database, we quantify the incidence rate of arrhythmia in patients and how this changes depending on patient characteristics. First, we propose that such datasets are a complementary resource for determining relative drug risk and assessing the performance of cardiac safety models for regulatory use. Second, the results suggest important determinants for appropriate stratification of patients and evaluation of additional drug risk in prescribing and clinical support algorithms and for precision health.

In vitro data and computational models can assist with calculating pro-arrhythmic risk

We use patient health records and FDA Adverse Event Reporting System reports

Use of such datasets helps assess relative drug risk and cardiac safety models

We quantify how patient characteristics can affect arrhythmia incidence

Validation of broad panel clinical sequencing-based genomic risk stratification in patients with advanced lung adenocarcinomas

9113

Background: We recently established the ability of broad-panel clinical sequencing data to stratify overall survival of patients with advanced lung adenocarcinomas in a single institutional experience (Shen, Riely et al., JCO Precision Oncology 2019). Here we sought to assess its generalizability to a broader range of patients (including patients from multiple community and academic sites) using a different sequencing panel, with an integrated electronic health record and genomic database. Methods: We identified 2,779 next-generation sequencing-tested patients with advanced lung adenocarcinomas from the Flatiron-Foundation Medicine Clinico-Genomic database. A genomic risk model developed from the initial discovery cohort (n=1,054) was used to calculate a risk score for each patient in the validation cohort, scaled between 0 and 10, indicating the risk of cancer specific mortality. Results: Patients in the validation cohort were classified into four risk categories with median survival ranging from 37.6 months (95% CI: 32.9-43.8) in the low risk group (n=534) to 10.9 months (95% CI: 8.0-16.5) in the highest risk group (n=75), representing a hazard ratio of 3.0 (95% CI: 2.2-4.1) and closely matching the discovery cohort observations. A smaller proportion of patients were deemed high risk in the validation cohort (2.7% vs 10% in the discovery cohort). There were some differences in the frequencies of the most common genomic alterations between the validation and discovery cohorts, including TP53 (57.3% vs 55.1%) , KRAS (32.8% vs 30%), EGFR (18.6% vs 29.4%) as well as overlapping STK11 and KEAP1 co-mutations (2.4% vs 10%). Conclusions: We demonstrate that a clinical tumor sequencing-based genomic risk stratification strategy can be applied broadly across cohorts and different sequencing panels and platforms, to improve the understanding of heterogeneity in clinical outcome for patients with metastatic lung adenocarcinomas and the mutation and co-mutational patterns that underlie such heterogeneity.

Time-to-treatment discontinuation (TTD) and real-world progression-free survival (rwPFS) as endpoints for comparative efficacy analysis between entrectinib trial and crizotinib real-world ROS1 fusion-positive (ROS1+) NSCLC patients

9070

Background: Entrectinib is an oral tyrosine kinase inhibitor for ROS1+ NSCLC. Three phase 1/2 single-arm studies showed entrectinib efficacy in this population (Doebele WCLC 2018). Due to the rarity of ROS1+ pts generating direct comparative evidence in prospective randomized trials is difficult. We identified a retrospective real-world cohort of ROS1+ NSCLC pts from electronic health records (EHR), to compare crizotinib, the current standard of care, to entrectinib as reported in clinical trials. Methods: Crizotinib-treated pts with advanced ROS1+ NSCLC diagnosed 1 Jan 2011 to 30 Jun 2018, were identified with technology-enabled abstraction in the Flatiron Health EHR-derived database ( > 2.1 million cancer pts from US oncology practice). Entrectinib trial inclusion/exclusion criteria were applied to match the crizotinib cohort as closely as possible. Primary endpoint: TTD, adapted from Gong (ASCO 2018); rwPFS (physician/scan report) and OS were secondary outcomes. Time-to-event analyses used Kaplan-Meier survival curves and Cox proportional hazard models on propensity score weighted populations; age, gender, race/ethnicity, smoking status, brain metastasis and previous lines of therapy were prognostic factors. Results: We analyzed 53 entrectinib and 69 crizotinib ROS1+ NSCLC pts. Median weighted TTD: entrectinib, 14.6 mo (95% CI: 8.3–23.8); crizotinib, 8.8 mo (95% CI: 8.2–9.9). When rwPFS from crizotinib was compared to trial PFS, entrectinib had longer PFS vs crizotinib (weighted HR: 0.44; 95% CI: 0.27–0.74). Median OS with entrectinib was not reached (median follow-up: 15.5 mo); weighted median OS with crizotinib was 18.5 mo (95% CI: 15.1–19.9). Findings were consistent across multiple sensitivity analyses. Conclusions: Entrectinib was associated with longer TTD and PFS in ROS1+ NSCLC pts vs a matched real-world crizotinib population. Control populations derived from real-world cohorts can supplement evidence from clinical trials in settings where new standards of care are needed, but where only limited data are available and randomization is not feasible.

Real-world anaplastic lymphoma kinase (ALK) rearrangement testing patterns, treatment sequences, and survival of ALK inhibitor-treated patients

BACKGROUND

The anaplastic lymphoma kinase (ALK) treatment landscape is crowded following recent ALK inhibitor approvals, and updated information on real-world treatment patterns in advanced non-small-cell lung cancer (aNSCLC) with ALK rearrangement (ALK+) is needed.

METHODS

This retrospective US cohort study used Flatiron Health's longitudinal electronic health record (EHR)-derived database. Patients (≥ 18 years old) diagnosed with stage IIIB/IV aNSCLC, with documented ALK rearrangement and ≥2 visits after January 1, 2011 were followed until February 28, 2016. Patients enrolled on a clinical trial or exposed to ALK inhibitors other than crizotinib or ceritinib were excluded. Treatment patterns, time and type of biomarker testing, and overall survival (OS) were analyzed.

RESULTS

Median age (n = 300) was 62.5 years; 55% female; 48% non-smokers; 8.7% central nervous system (CNS) metastases at diagnosis. Overall, 73% and 86% received their first ALK biomarker test before/at diagnosis, or before/during first-line treatment, respectively. In total, 90.0%, 78.1%, and 74.7% received first-, second-, and third-line therapy, respectively. Most patients received ALK-targeted treatment; 62% received crizotinib, of which 21% reported a dose reduction. Progression was the most common reason for crizotinib (78%) and ceritinib (41%) discontinuation. Median OS was 29.4 months (95% CI =24.7-39.6) overall; 27.1 months (95% CI =22.0-35.0) in patients with CNS metastases, and 36.9 months (95% CI =25.1-not reached) without.

CONCLUSIONS

Despite widespread crizotinib use in patients with ALK+ aNSCLC, a high proportion of patients progressed. Ongoing analyses of EHR-derived cohorts are valuable in assessing real-world testing rates and therapeutic use of ALK inhibitors.

Comparative effectiveness from a single-arm trial and real-world data: alectinib versus ceritinib

Aim: To compare the overall survival of anaplastic lymphoma kinase-positive non-small-cell lung cancer patients who received alectinib with those who received ceritinib. Materials & methods: Two treatment arms (alectinib [n = 183] and ceritinib [n = 67]) were extracted from clinical trials and an electronic health record database, respectively. Propensity scores were applied to balance baseline characteristics. Kaplan–Meier and multivariate Cox regression were conducted. Results: After propensity score adjustment, baseline characteristics were balanced. Alectinib had a prolonged median overall survival (alectinib = 24.3 months and ceritinib = 15.6 months) and lower risk of death (hazard ratio: 0.65; 95% CI: 0.48–0.88). Conclusion: Alectinib was associated with prolonged overall survival versus ceritinib, which is consistent with efficacy evidence from clinical trials.

Reduction potential of microbial, odour and ammonia emissions from a pig facility by biofilters

The intention of this study was the determination of the potential to reduce specific microbial bioaerosol (cultivable bacteria and fungi, total cell counts of microbes, airborne endotoxins and microbial volatile organic compounds, MVOC), odour and ammonia emissions from a pig facility by biofilters. Five identical biofilter units in half technical scale were filled with different filter materials (Biochips, coconut-peat, wood-bark, pellets + bark and compost) and connected in parallel to a piggery. The results showed obvious differences between the filter materials. Numbers of airborne cultivable bacteria were decreased by ca. 70 to 95% and the total counts of bacterial cells from ca. 25 to (>) 90%. The total amount of fungal cells was reduced by at least 60%, although the percentage of cultivable moulds in the air after passing the filters was sometimes higher than before. Airborne endotoxins and MVOC were effectively reduced by all filter materials to at least 90%. Regarding odour, the average reduction was between 40 and 83%, whereas only one of the filters proved to be capable of slightly reducing the ammonia emissions. No relationships between odour/ammonia and microbial bioaerosols with regard to the reduction efficiency of the different filter materials or the total load of the emitted air could be established. A tendency could be shown, that biofilters best capable to reduce odour emitted slightly more airborne bacteria, both cultivable and total cell counts.

Biofilters--the influence of different filter materials and different operating conditions on the reduction efficiency

A series of biofilter materials were tested for performance efficiency in treating ventilation air from pig sheds. Two different phases were conducted. In phase A five different biofilter materials were tested with the aim of selecting the best material regarding the odor reduction and over all efficiency. In conclusion of phase A, biochips, a new filter material, was selected due to its high average odor reduction (81%) in combination with a very low flow resistance (less than 30 Pa for filter volume loads of 600 m3 h(-1) m(-3)). In phase B, biochips and coconut fiber peat were tested under different operating conditions. Biofilters number 3 and 4 each with down-flow air supply could realize a higher average odor reduction than biofilter 5 with up-flow air supply. All three filter materials had the same moisture, however the two biofilters with down-flow air supply showed a more homogenous moisture distribution but they also used approximately three times more water than biofilter 5. Like in phase A the specific odor cleaning efficiency was mainly influenced by the specific odor loading rate and showed only slight differences between the five biofilters differing in bulk layer height (0.5 m and 1.0 m) and air supply mode. All in all the investigation showed that the efficiency of biofilters can be optimized by using an new filter material and a appropiate operating mode.