Research: Publications highlighting emerging technologies, pre-commercial research, and investments for the innovations of tomorrow
Behavioral Health
Psychiatric Annals
In this article, authors review the history of natural language processing, provide examples of its application in psychiatry, and address future direction and potential applications in health and healthcare delivery.
Oncology
Cell
Tumor ecosystems are comprised of cancer cells plus other cells and structural components the cancer interacts with such as immune cells. The authors used single-cell mass cytometry of 144 human breast cancer samples compared to 50 non-tumor tissue samples to study the cellular make up and physical expression of the genetic makeup of these tumor ecosystems. Tumors displayed individuality in ecosystem cell composition. Relationship analyses between tumor and immune cells revealed characteristics of ecosystems related to immunosuppression and poor prognosis.
Real World Evidence: Publications featuring our unique data assets and industry-leading scientists
Behavioral Health
Academy Health Annual Research Meeting
As we continue to seek solutions to the opioid crisis domestically and abroad, a new study was able to identify characteristics associated in counties with high vs. low rates of opioid-related hospitalization rates (classified as hot spots). This study measures several characteristics in the evaluation of hospital discharge data for 45 states and the District of Columbia, indicating the potential need to target opioid-related interventions to local community characteristics.
Pain Medicine
Patients who were adherent to pain management guidelines experienced significantly lower health care resource use and costs, such as fewer emergency department visits, compared to non adherent patients in a retrospective cohort study of IBM® MarketScan® claims data (n=682,802).
Academy Health Annual Research Meeting
Employees with opioid use disorder (OUD) not receiving pharmacotherapy had approximately $19,182 more healthcare costs than other employees in a study evaluating claims data for a sample of Fortune 500 companies from the IBM® MarketScan® Research Databases. In comparison, employees with OUD that received pharmacotherapy had approximately $5,460 more in healthcare costs than other employees, supporting a case for expanding access to pharmacotherapy treatment in OUD.
Academy Health Annual Research Meeting
Spending for mental and substance use disorder (SUD) treatment rose 62% between 2006 and 2015 ($131 to $212 billion) in an analysis of data from the Centers for Medicare & Medicaid Services and IBM® MarketScan® Research Database. This increase outpaced growth in overall healthcare spending. A more pronounced increase seen in SUD spending coincided with the growth in the opioid epidemic and accelerated following passage of federal mental health parity legislation and Medicaid expansion in some states.
JAMA Network Open***
As opioid-related mortality continues to increase, the opioid overdose reversal drug naloxone remains a critical intervention in preventing overdose death. New data indicates that prescribing in emergency, inpatient, and outpatient settings represents an opportunity to improve access to naloxone therapy. Overall, 98.5% (n = 135,973) of high-risk patients did not receive naloxone, despite many interactions with the health care system in an evaluation of claims data from IBM® MarketScan® Research Databases.
Academy Health Annual Research Meeting
More mental health inpatient stays, partial shortages of mental health professionals, higher unemployment and poverty rates, and a higher percentage White population were associated with higher county-level rates of neonatal abstinence syndrome (NAS) in an evaluation of data from 2,439 counties in 47 states, representing 96.2% of the US population. Counties with the highest rates of NAS were concentrated throughout Appalachia, Maine, and Vermont, and in areas of the Great Lakes, Colorado, Montana, New Mexico, and Utah.
Healthcare Reform
Health Services Research
Rates of uninsured inpatient discharges and emergency department visits fell in many demographic groups across Medicaid expansion states, while rates remained stable or increased slightly in nonexpansion states, in an evaluation of data from the Healthcare Cost and Utilization Project.
Immunologic Diseases
Pharmacoeconomic Open
Patients with rheumatoid or psoriatic arthritis covered by insurance plans with step therapy requirements had lower odds of treatment effectiveness compared with patients in plans without access restrictions or with prior authorization only in an evaluation of IBM® MarketScan® claims data (n=5706). This difference was primarily due to lower odds of adherence over the 12 months following initiation of subcutaneous biologic disease-modifying antirheumatic drugs.
Drug Information Association 2019 Global Annual Meeting
Application of disease-model definitions and a visual analytics engine applied to IBM® MarketScan® Research Databases identified differences in outcomes and cost associated with rheumatoid arthritis patients without the need for data science, clinical or programing knowledge.
Metabolic Diseases
Diabetes***
The total annual payment for insulin increased from $1,982 in 2003 to $6,250 in 2016, with a 3-fold annual rate increase seen between 2013 and 2016 in an evaluation of IBM® MarketScan® claims data for adult patients with diabetes (n=1,703,343). While out-of-pocket payments increased as well, analyses indicate that payers bore most of the increase.
Oncology
Journal of Clinical Oncology***
The new CMS oncology outcomes measure (OP-35) defines 30-day post chemotherapy inpatient and emergency room utilization as “potentially avoidable” if the events involved nausea, vomiting, or one of eight other toxicities. Analysis of clinical data from the IBM Watson Explorys database of oxaliplatin treatment for colorectal cancer (n=4,231 courses of oxaliplatin) found that approximately one third were associated with “potentially avoidable” nausea and/or vomiting, indicating a need for more aggressive preventative therapy.
Journal of Clinical Oncology***
Mean total cost of care and total drug cost per patient for cancer patients over a 21-month period were $141,415 and $13,579, respectively, in an evaluation of IBM® MarketScan® claims data. Total pharmacy expenditures across all study patients (n=195,290) were approximately $2.5 billion, with antineoplastic drugs making up the largest portion. The potential for unsustainable growth in the cost of cancer care support the need for investigating drug price regulation.
Renal
American Journal of Kidney Diseases: the Official Journal of the National Kidney Foundation
A model applying machine learning to real world data accurately classified risk of end-stage kidney disease or 50% reduction in kidney function for patients within 5 years of a diagnosis of immunoglobulin A nephropathy [n=2047; C statistic of 0.84 (95% CI, 0.80-0.88)].
Real World Evidence and Artificial Intelligence Combined: Studies applying artificial intelligence technologies, such as machine learning, to deliver more robust insights from real world data.
Oncology
Radiology
A deep learning algorithm was trained on 9,611 mammogram images linked to electronic health record data to make breast cancer predictions on biopsy malignancy and differentiation between normal and abnormal screening. For malignancy prediction, the area under the receiver operating characteristic (ROC) curve was 0.91 (95% CI: 0.89, 0.93)
with specificity of 77.3% (95% CI: 69.2%, 85.4%) at a sensitivity of 87%.
Area under the curve of ROC is a measure of how well a model distinguishes between two groups
, indicating potential utility for this model to assist radiologists in breast cancer diagnosis.
Artificial Intelligence: Publications featuring our AI-enabled solutions
Metabolic Diseases
Diabetes
A machine learning model achieved 90% accuracy in predicting hypoglycemic events within a two-hour window and 85% accuracy in predicting hypoglycemic events within a four-hour window, with anonymized data from approximately 10,000 patients using Medtronic Guardian™ Connect CGM and Medtronic MiniMed™ 530G.
Oncology
Journal of Clinical Oncology***
A multidisciplinary tumor board (MDT) reviewed and ultimately chose treatments they had not previously considered, based on information from Watson for Oncology, for 13.6% of cases after reviewing information on options for 1,000 cases of breast, lung and colorectal cancer in this blinded study.
Journal of Clinical Oncology
Over 85% of Brazilian oncologists surveyed (n=7; 903 cases) agreed that Watson for Oncology (WfO) provided relevant and actionable information at the right time in their workflow. Additionally, over 50% shared that WfO exceeded their expectations as a clinical decision support tool for managing patients.
Journal of Clinical Oncology***
Patients receiving consultation with a 7-step approach supported by Watson for Oncology indicated higher satisfaction toward treatment options, higher confidence level in their health care workers, and greater willingness to follow the treatment option when compared to patients in a traditional non-CDS system group.
Journal of Clinical Oncology
In interpreting 54 South Korean patients with blood-based cancers, Watson for Genomics' annotations of sequencing results correlated well with manually curated expert opinion (90% in randomized subset) and identified clinically actionable insights missed by manual interpretation (33%).
Annals of Oncology
Watson for Clinical Trial Matching enabled high-volume patient screening in less than half the time compared to manual efforts in the Mayo Clinic Lung Oncology practice.
Journal of Clinical Oncology
A study comparing treatment options from Watson for Oncology (WfO) to shared treatment decisions made in 48 Brazilian patients with prostate cancer found complete concordance in 54% of cases and partial concordance in 15% of cases. While great variation exists in the treatment of prostate cancer, WfO was at least partially concordant with shared decision making in the majority of patients.
Journal of Clinical Oncology
A machine learning model was 93% accurate in classifying high-quality, clinically relevant papers from a set of 988 PubMed abstracts related to cancer treatment when applied to evaluate the quality of publications.
Journal of Clinical Oncology***
More than 80% of primary hospitals surveyed (n=56) were willing to learn about standardized treatments and recent treatment progress of tumors via remote consulting augmented with Watson for Oncology, indicating an interest in applying the technology to better understand standard treatment and progress on the treatment of cancer.