Correspondence on “A common longitudinal intensive care unit data format (CLIF) for critical illness research”. Author’s reply

We thank Popoff et al. [1] for their thoughtful commentary on our Common Longitudinal ICU data Format (CLIF) [2] and appreciate their suggestions for improving CLIF.

First, we agree that integrating CLIF with standard terminologies such as Logical Observation Identifiers Names and Codes (LOINC) and RxNorm is necessary for CLIF’s scalability and the extraction of real-time CLIF tables with HL7 FHIR. CLIF uses a backward-compatible versioning system to standardize a clinician-curated set of minimal common ICU data elements (mCIDE) that were reliably identified at each founding institution. We view integration with existing terminologies not as replacing our current minimum Common ICU Data Elements (denoted by *_category variables), but as a mapping project. For example, clinical drug RxCUIs 242969 and 569844 both map to “norepinephrine” in the CLIF continuous medication table.

Second, we acknowledge that “long” tables are a more computationally efficient storage structure and most CLIF tables use this format (e.g., vitals, labs, medications). The “wide” CLIF tables are limited to those that describe life support device treatments where settings should be documented near-simultaneously and in specific patterns (respiratory support, ECMO/MCS, and CRRT). For example, a patient receiving “assist control-volume control” mode of mechanical ventilation is expected to have a set tidal volume but not a set pressure control. Keeping the respiratory support table in a wide format ensures that clinical ICU domain knowledge is encoded directly into CLIF, making data quality issues or mapping errors easier to identify.

Third, we recognize that site-specific extraction, transformation, and loading (ETL) into CLIF requires considerable effort from participating institutions. However, as of June 2025, over 10 unique ICU data science teams across North America have adopted CLIF and contribute high-quality, granular data on over 800,000 ICU admissions for federated research [3]. This uptake demonstrates that CLIF is accessible and intuitive to ICU data scientists. By focusing on a small but precisely defined mCIDE, we are optimistic CLIF can continue to scale.

Fourth, CLIF is designed to complement, not compete with, the OMOP Common Data Model. We envision a generalizable pipeline from OMOP to CLIF using the standard terminologies discussed above. However, this approach will be expensive and time-intensive, given that 500 h of highly skilled data scientist effort were required to map only 64% of MIMIC data (a single healthcare system) into the OMOP CDM [4]. OMOP also currently lacks standard concepts for many ICU data elements [5].

Finally, INDICATE is an inspiring project with exciting aims, and we applaud the European Union for funding this ambitious effort. The CLIF consortium does not yet have direct funding; rather, CLIF has been “crowd-sourced” by an organically grown group of researchers motivated to use CLIF for their individual research projects. Therefore, we view CLIF as complementary to INDICATE and other significant ICU data science investments [6]. We will continue to expand the open-source mCIDE and CLIF code ecosystem over time, and hope that CLIF becomes an appealing research format for INDICATE researchers and ICU data scientists worldwide.