Validation of Novel Identification Algorithms for Nonfatal Myocardial Infarction Using Uniform Objective Criteria and Pooled Clinical Trial Data

Author(s)

Ports K¹, Chen W², Saunders D², Koduru B², Aptekar J², Spanias J², Strait C², Tso C², Buderi R², Gerome D³, Bourlon PL², Saltzman S⁴, Jain T², Talwai A², Jain R²
¹Medidata AI, Medidata a Dassault Systèmes company, San Diego, CA, USA, ²Medidata AI, Medidata a Dassault Systèmes company, New York, NY, USA, ³Medidata AI, Medidata a Dassault Systèmes company, Huntington, NY, USA, ⁴Medidata AI, Medidata a Dassault Systèmes company, Boston, MA, USA

Presentation Documents

ISPOR23_Ports_POSTER126335.pdf

OBJECTIVES:

Nonfatal myocardial infarction (MI) is primarily identified using diagnostic code-based algorithms in real-world data (RWD). This investigation aimed to utilize pooled clinical trial (CT) data to identify clinical indicators to optimize RWD MI identification.

METHODS:

Anonymized historical CT data from the Medidata Enterprise Data Store was pooled on patients over 40 with established cardiovascular (CV) disease or CV risk factors. Algorithms were adapted from the MI identification criteria suggested by the Standardized Data Collection for Cardiovascular Trials Initiative (SCTI) and FDA and included only the information available in RWD. Algorithm-1 (A1) only included cardiac biomarker assessment and was augmented in additional algorithms by adding different combinations of ECG assessment, signs and symptoms, treatments, and all-cause hospitalization occurring within predefined timeframes (A2-A7). Using the clinical events committee adjudicated events as the gold standard, algorithm accuracy was assessed using positive predictive value (PPV) and sensitivity.

RESULTS:

A total of 40,866 patients were included in the analysis (median follow-up of 1.5 years), with 1133 adjudicated MI events. Cardiac biomarker assessment alone (A1) produced the highest sensitivity (99%) but low PPV (27%). The addition of ECG and hospitalization increased PPV (38%) with a slight reduction in sensitivity (95%). Algorithms that included signs and symptoms (sensitivity = 22%) and treatments (sensitivity = 66%) produced the lowest sensitivities without meaningful gains in PPV.

CONCLUSIONS:

High algorithm sensitivities indicate the validity of specific clinical indicators in MI identification. Low PPVs may be an artifact of not including test results and reflect confounding morbidities with similar diagnostic and treatment interventions. Including hospital discharge diagnoses, not available in CT data, may increase algorithm PPV when utilized with RWD. These findings can be incorporated in RW algorithms to identify MI in retrospective studies or prospectively in pragmatic trials and highlight pooled CT data's utility in identifying RW clinical events.

Conference/Value in Health Info

2023-05, ISPOR 2023, Boston, MA, USA

Value in Health, Volume 26, Issue 6, S2 (June 2023)

Code

EPH4

Disease

Cardiovascular Disorders (including MI, Stroke, Circulatory), No Additional Disease & Conditions/Specialized Treatment Areas

Presentation