|
Abstract Introduction: Data review in a high-throughput clinical setting has traditionally involved the monotonous evaluation of every chromatogram at the expense of employees’ sanity. This review is as much art critique as data science, given the confidence to release a patient result more tangibly comes from placing eyes on picture of a chromatogram rather than a set of metrics. However, those eyes are looking for different things depending upon the reviewer, assay, day, week, month, and year, and may lead to inconsistency in data review and patient care. Further, the wealth of data from each patient that could provide insight into everything from instrument capacity to assay performance to personnel utilization remains largely unexplored.
To tackle these challenges, we employ the cloud-based software Ascent® (Indigo Bioautomation) to dynamically model peaks and apply mathematical rules to flag samples based on acceptance criteria for patient result release or repeat/re-extraction. After batch release, results are transferred to a secure Amazon Redshift Server, where the data is accessed for analysis using the R statistical programming language. This talk will describe our journey into the world of big data, what we have learned and where we want to go.
Methods: Once batch acquisition is complete, sample information from the batch is merged with assay-specific lot and level information and transferred to a SSAE-16 (HIPAA, SOX, PCI-DDS) compliant, secure data center with advanced physical and logical security controls provided by Amazon Web Services (AWS). The data is accessed on a desktop application using the RPostgreSQL package that allows seamless introduction into the Tidyverse library of packages (dplyr, ggplot2, forcats, lubridate, etc).
Results: Our clinical mass spectrometry lab processes ~500 batches a week across 29 separate assays and 21 instruments. This corresponds to up to 24 employees reviewing daily > 8000 unique samples, many of which have 2-4 chromatograms per sample per analyte, generating a dizzying array of data. To translate this data into a form that can be used to monitor and improve the quality and efficiency of our laboratory, we use three tools: (1) Dynamic queries to the AWS database by situation (2) automatic, static reports generated weekly and monthly and (3) interactive dashboards constructed using R/Shiny for dynamic analysis. This talk will focus on how we use these tools to answer questions like the following: Do data review team members avoid reviewing batches that have a high(er) flag rate (samples with problems?) How long does it take to review a batch once it is ready? Are instruments being used at their full capacity? How can we use this data to improve throughput? What do assay flags tell us about the health of an assay or instrument? How can we concisely view Calibrator and QC performance across 29 separate assays and multiple levels daily, weekly, and monthly? |
= Discovery stage.
= Translation stage.
= Clinically available.
