Mapping from the International Classification of Diseases (ICD) 9th to 10th Revision for Research in Biologics and Biosimilars Using Administrative Healthcare Data

Abstract

Purpose: The Centers for Medicare and Medicaid Services (CMS) mandated the transition from ICD-9 to ICD-10 codes on October 1, 2015. Postmarketing surveillance of newly marketed drugs, including novel biologics and biosimilars, requires a robust approach to convert ICD-9 to ICD-10 codes for study variables. We examined three mapping methods for health conditions (HCs) of interest to the Biologics and Biosimilars Collective Intelligence Consortium (BBCIC) and compared their prevalence. Methods: Using CMS General Equivalence Mappings, we applied forwardbackward mapping (FBM) to 108 HCs and secondary mapping (SM) and tertiary mapping (TM) to seven preselected HCs. A physician reviewed the mapped ICD-10 codes. The prevalence of the 108 HCs defined by ICD-9 versus ICD-10 codes was examined in BBCIC's distributed research network (September 1, 2012 to March 31, 2018). We visually assessed prevalence trends of these HCs and applied a threshold of 20% level change in ICD-9 versus ICD-10 prevalence. Results: Nearly four times more ICD-10 codes were mapped by SM and TM than FBM, but most were irrelevant or nonspecific. For conditions like myocardial infarction, SM or TM did not generate additional ICD-10 codes. Through visual inspection, one-fifth of the HCs had inconsistent ICD-9 versus ICD-10 prevalence trends. 13% of HCs had a level change greater than +/−20%. Conclusion: FBM is generally the most efficient way to convert ICD-9 to ICD-10 codes, yet manual review of converted ICD-10 codes is recommended even for FBM. The lack of existing guidance to compare the performance of ICD-9 with ICD-10 codes led to challenges in empirically determining the quality of conversions.

Health outcomes coding trends in the US Food and Drug Administration's Sentinel System during transition to International Classification of Diseases‐10 coding system: A brief review

“…Our findings suggest that all studies using ICD‐9‐CM‐based algorithms mapped to ICD‐10‐CM codes should assess the performance of the mapping and undertake detailed reviews before using them. Although prior studies found that FBM may be generally more sensitive compared to SFM, 2,9‐11 our findings argue for careful review of all mapping approaches. Notably, a mapping method that is generally better than others may not be the best for all HOIs.…”

“…Mapping methods should be customized for individual studies and specific HOIs. Also, consistency in incidence and prevalence trends may be influenced by multiple factors, including parameters in study design, such as cohort re-entry conditions, length of wash-out period, care settings examined Furthermore, additional analyses of these reports would enhance our understanding of the value of manual code curation, as recommended by a prior study that reported manual refinement of a mapped ICD-10 algorithm of meningitis improved the consistency in prevalence, 11 and the best methods for assessing changes in incidence and prevalence across the ICD-9-CM and ICD-10-CM eras. A more in-depth analysis of the existing reports, or future reports, would include analysis of monthly incidence and prevalence data to allow more systematic assessment of changes in level and trend (i.e., beyond visual inspection) and investigation of whether manual curation can improve HOI phenotype performance as compared to prior trends.…”

Risk of cardiovascular events in patients having had acute calcium pyrophosphate crystal arthritis

“…Non-fatal CV event and death were prespecified secondary outcomes. We employed validated algorithms to identify non-fatal CV event using International Classification of Diseases ninth edition (ICD-9) diagnosis codes and procedural codes, updated to include corresponding ICD-10 codes 17–20. These algorithms defined MI, ACS and stroke using inpatient diagnosis codes.…”

“…Comorbidities previously associated with risk of CV disease or death, including diabetes, hypertension, hyperlipidaemia, atrial fibrillation and cancer, were defined by ≥2 ICD-9/10 codes from EHR entry through index date (see table 1 for a complete list of CV risk factors) 20. Requiring at least two ICD codes increases PPV for each comorbidity compared with one code; assessing comorbidities in all available data before index date is not expected to create bias 22 23.…”

Acute myocardial infarction: Development and application of an ICD-10-CM-based algorithm to a large U.S. healthcare claims-based database

“…This process involves translating ICD-9-CM codes to ICD-10-CM (forward-mapping), then translating the identified ICD-10-CM codes back to ICD-9-CM (backward-mapping) to capture codes that may not have one-to-one mapping between versions and to identify additional related codes that may be of relevance. Recent publications have reported that this approach is appropriate for efforts to define AMI using ICD codes [7,8]. The draft algorithm builds on these prior studies, using a more current version of the ICD-10-CM and an approach informed by high-quality ICD-9-CM validation studies, and was subject to further review and refinement by clinical subject matter experts.…”