Writing for Pediatric Critical Care Medicine: The Single-Center Retrospective Cohort Study

Abstract

In the 30 months from January 2023 to June 2025, Pediatric Critical Care Medicine (PCCM) published the results of approximately two single-center retrospective cohort studies per monthly issue. That is, out of 350 clinical research articles published in this period, one-in-six is single-center retrospective work. Clearly, PCCM remains a destination for this type of research output, even when other journals no longer consider such works for publication. We welcome this content, but now the competition for space in PCCM is getting harder and the needs of readers are changing. Therefore, the second item this year in the “Writing for PCCM” series (1) is aimed at helping researchers and authors to better understand the new direction of the Journal when assessing single-center retrospective cohort studies. WHAT TYPES OF STUDIES ARE IN THIS CATEGORY? The exposure of interest in a single-center retrospective cohort study has already occurred in an at-risk population. Often, the data being analyzed were originally collected or stored for reasons other than the question being considered by the researchers (e.g., in physician notes, laboratory test results, or diagnostic test reports). Hence the research methodology needs to be rigorous if the work is to be meaningful, and it usually follows a series of steps (2,3). For example: Step 1, conceiving an idea for a research question; Step 2, carrying out a background literature review; Step 3, writing a detailed research proposal; Step 4, constructing the tools for data abstraction and analysis; Step 5, developing protocols and operating procedures for data handling; Step 6, calculating sample size and assessing potential analytic strategies for dealing with missing data; Step 7, gaining institutional research ethics approval; Step 8, abstracting data; Step 9, conducting a pilot study; and Step 10, review and moving on to the full study. Our field is broad and multidisciplinary, and so it should be no surprise that PCCM authors (January 2023 to June 2025) have presented a range of topics using single-center retrospective cohorts. Five areas of interest include: 1) descriptions of international relevance (e.g., Dengue, sepsis phenotypes, and healthcare practices in different settings); 2) summaries of organizational or administrative processes in the PICU such as electronic health record monitoring, clinical service models, diagnostics using large language models, and development of complications and mortality models; 3) a focus on organ system-oriented specialty practices (e.g., cardiac, neurology, pulmonology, general medicine and hematology/oncology); 4) an analysis of treatment-oriented practices (e.g., fluids and extracorporeal renal or heart-lung support); and 5) care at the end-of-life (e.g., trajectory or pathway of palliation, clinical testing and ancillary studies for the determination of death using neurological criteria, and family-centered care). The question looking beyond 2025 is whether the above broad and somewhat uncoordinated and idiosyncratic approaches to research topic selection are useful to readers and important for the future of our field? In response to such a criticism, our new modus operandi (Latin meaning “particular way of doing something”) at PCCM is to support single-center retrospective cohort studies which, in the writing, emphasize one of three key themes in the research: 1) describing some aspect of clinical care that is new and unique; or 2) modeling healthcare-related outcomes using new approaches with data; or 3) generating hypotheses that contribute to a larger program of research or narrative within PCCM. The sections below provide some context and examples of these new frameworks for prospective PCCM authors. DESCRIBING AN ASPECT OF CLINICAL CARE THAT IS NEW AND UNIQUE Consider the 2023 updated consensus guidelines on pediatric brain death/death by neurologic criteria (BD/DNC) published by national organizations in Canada (4) and the United States (5). In the update from Canada, but not in the U.S. guidance, there was an important change in the use of radionuclide scintigraphy as an ancillary test in infants under 2 months of age (4). That is, there was a conditional recommendation against the use of ancillary testing in under 2-month-olds, which was based on a systematic review (6) and case reviews (7) showing a paucity of direct evidence on diagnostic certainty. In response, in 2024, a group from a single-center, quaternary, academic children’s hospital in the United States reported their unique experience of ancillary testing in the evaluation of BD/DNC over the period 2005 to 2022 (8). The retrospective report described 10 infants aged under 1 year whose evaluation for BD/DNC included a nuclear medicine cerebral perfusion ancillary test, and 9/10 studies were consistent with BD/DNC. This single-center retrospective cohort study was timely, spoke to a contemporary clinical need, and informed the field; it has also received correspondence (9,10). Another example of new and/or unique care described in a single-center retrospective cohort study is the 20-year experience (1998 to 2017) of mental health provision in a PICU using an embedded psychologist for referrals (published in 2023) (11). There was an accompanying editorial because this descriptive work supported PCCM’s theme on post-intensive care syndrome (12). MODELING HEALTHCARE-RELATED OUTCOMES USING NEW APPROACHES WITH DATA Understanding the why, what and how of outcomes in patients “will continue to be a cornerstone of pediatric critical care” (13). In 2024, the Pediatric Data Science and Analytic (PEDAL) subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) network provided PCCM with a 5-year roadmap for outcomes research using machine learning-based clinical prediction models (13). As such, a key tenet of the PEDAL-PALISI subgroup is work with “an early focus on implementation (14), preferably in coordination with experts in implementation to ensure...accurate models lead to improved clinical care.” In other words, progress in critical illness healthcare-related outcomes research is brought about by bringing together data science, modeling, external validation, and implementation science (13–16). This second emphasis for single-center retrospective cohort studies in PCCM has the potential to be a huge inter-disciplinary endeavor. As an example of part of this journey, consider the paired articles about dynamic prediction of mortality in the PICU that were published in 2024 (17,18), along with an accompanying editorial (19). Researchers in a single-center compared admission-day pediatric Sequential Organ Failure Assessment (pSOFA) score with longitudinally measured pSOFA scores in their respective accuracy of mortality prediction in a 7-year cohort of over 9,000 patients (18). The investigators reported that in their PICU the longitudinal scores improved the accuracy of mortality prediction. To achieve this dynamic modeling, the researchers first developed an automated, open-source, scalable method for calculating pSOFA directly from the electronic health record (17). Further scaling to multiple centers, and analysis of how implementation improves PICU practice are the next evolutions that would complete this research journey. PCCM does not expect data research reports to provide the complete work. However, the direction and purpose of the investigation using a single-center retrospective cohort needs to be explained, which takes us to the third emphasis for writing-up these studies for PCCM. GENERATING HYPOTHESES THAT CONTRIBUTES TO A LARGER RESEARCH PROGRAM A phrase that often appears at the end of a research report in PCCM reads something like “…and future studies should explore….” These words risk being overlooked or even ignored, but they are relevant to the third “frame” for single-center retrospective cohort studies in PCCM. That is, PCCM is seeking single-center retrospective cohort studies that extend a larger research program or narrative. For example, authors of a single-center retrospective cohort (2014 to 2017) study found that greater family presence on the PICU was associated with lower odds of delirium in patients (20). The authors concluded that because “family presence is a modifiable factor that may mitigate the burden of pediatric delirium…future studies should explore barriers and facilitators of family presence in the PICU.” Of note, this work published in April 2025 represented progression from the authors’ previous single-center retrospective cohort (2011 to 2017) study on family presence in the PICU published in 2023 (21). Family-centered care is a new narrative theme within PCCM (22), which incorporates facilitating family presence on the PICU. Pediatric delirium is another theme within PCCM, with publication of multiple reports, systematic reviews, and the 2022 Society of Critical Care Medicine clinical practice guidelines on prevention (23). Thus, the third emphasis for single-center retrospective cohort studies in PCCM is a focus on extending the PCCM research narrative by anchoring new reports on what has preceded, and by bringing this development to the attention of readers. The report should also suggest progression: for example, the April 2025 report guided readers to explore “barriers and facilitators of family presence in the PICU” (20). We welcome learning about this next iteration in patient care. CLOSING COMMENTS One-in-six clinical research articles published in PCCM are reports using single-center retrospective cohort studies. This format has validity, but the emphasis and focus must change to meet the requirements of readers and our clinical community. PCCM therefore seeks reports that can be framed as either describing some aspects of clinical care that is new and unique; or modeling healthcare-related outcomes using new approaches; or generating hypotheses that contribute to a larger program of research or narrative within PCCM. Finally, regarding writing, the format of the 3,000-word structured clinical research report (24) is suitable, albeit with four minor caveats. The summary of demographic data by group in Table 1 (see details in [24]) in any report does not require p values. Percentages are not used in place of proportions when the denominator is < 50. Differences between proportions can be summarized as mean and 95% confidence interval of the difference in percentages. The language of association rather than causation should be used throughout.

Topics & Concepts