Reply to the message regarding machine learning to identify patients

Benjamin Skov Cas Hansen,1-3 Christina Leal Rodriguez,2 David Placido,2 Hans Christian Thorsen Mayer,2,4 Anna Bourse Nielsen2 Nicola Derian,5 Soren Bronak,2 Stiggrup Andersen1

1Clinical Pharmacy Unit, University Hospital Zealand, Roskilde, Denmark; 2NNF Center for Protein Research, University of Copenhagen, Copenhagen, Denmark; 3Department of Biostatistics, Department of Public Health, University of Copenhagen, Copenhagen, Denmark; 4Department of Intensive Care Medicine, University Hospital Copenhagen (Rigshospitalet), Copenhagen, Denmark; 5Data support and development, Region Zealand, Denmark

Correspondence: Benjamin Skov Kass Hansen, Department of Intensive Care, University Hospital Copenhagen-Rigshospitalt, Belgdamsvi 9, Copenhagen, 2100, Denmark, tel +45 60 19 68 01, e-mail [email protected]

See the original paper by Dr. Cass Hansen and colleagues

This was in response to the letter addressed to the editor

Dear Editor

We would like to thank Houlind et al for carefully reading our paper and feedback, but we found that they miss the mark in some calculations, considering the scope of our study. First, while does not actually cite specific sources for dosing recommendations, summaries of product characteristics (SPCs) are a major source of dosing guidelines and deviations for presumably occasional SPCs.1 Second, although many drugs already lack direct dose-reduction schemes, and the article could have been more explicit (see the fourth limitation, however; p. 221), we chose these renal risk drugs because of their simple rules for dose adjustment: including In that drugs that do not have direct operational instructions, such as opioids, would be incompatible with activating our results (i.e. making our results a ‘best case scenario’ in terms of inappropriate doses). Third, we set out not to verify the accuracy of but to examine the predictability of inappropriate drug doses according to these recommendations, assuming they are correct (as clinical staff would when following the same instructions). Fourth, we respectfully point out that the eGFR is <30 mL/min/1.73 m2 It was not a result in our analyzes and was not used to temporarily reclassify the severity of CKD, and that we used not the lowest eGFR but all eGFR values ​​in the follow-up period (to calculate time at risk), which should prevent continued dosing. In fact, we used eGFR ≤30 mL/min/1.73 m2 as one of the inclusion criteria (p. 214 in Kaas-Hansen et al2) and to operationalize the concept of inappropriate dosing (p. 214 and Figure 1 in Kaas-Hansen et al.2), which in turn serves as the basis for the five actual outcomes: >0, 1, ≥2, ≥3, and 5 inappropriate daily doses. Finally, in resting on both creatinine P and urine output,3 The discrepancy of the latter in routine clinical data such as ours likely caused significant misclassification of acute kidney injury (AKI), and clinical observations potentially indicative of AKI were not available in our data. These challenges combine with patients including patients with at least one rapid sedimentation rate ≤30 mL/min/1.73 m2 Between admission and indication (meaning that most eligible patients likely had some degree of CKD or AKI) it can be argued that the purpose of using acute renal insufficiency as an exclusion criterion in the sensitivity analysis.

However, we agree with two points raised by Houlind et al. First, it also reflects the conclusion (p. 221 in Kaas-Hansen2), in-silico results must demonstrate their value in future evaluations in the target clinical context, before any real clinical benefit can be claimed, and such endeavors must use challenging end points to the maximum extent possible. Second, repeating analyzes with absolute eGFRs and SPCs to activate the results could constitute an interesting alternative approach, and one that may have served our study in addition to sensitivity analysis.


SB reports ownership in Intomics A/S, Hoba Therapeutics ApS, Novo Nordisk A/S and Lundbeck A/S, and board membership management in Proscion A/S and Intomics A/S, out of this connection. All other authors reported no conflict of interest in this communication.


1. Hvad indeholder en præparatbeskrivelse? –; 2022. Available from: accessed 31 mayo 2022.

2. Kaas-Hansen BS, Leal Rodríguez C, Placido D et al. Using machine learning to identify patients at high risk of inappropriate drug doses in periods with impaired renal function. clin epidimol. 2022; 14:213-223. doi: 10.2147/CLEP.S344435

3. KDIGO Board Members. KDIGO Clinical Practice Manual for Acute Kidney Injury. Kidney Int Supplements. 2012; 2:279. doi: 10.1038/kisup.2012.3

This work has been published and licensed by Dove Medical Press Limited. The full terms of this license are available at and include the Creative Commons Attribution – Non Commercial (unported, v3.0) license. By accessing the work, you agree to the terms. Non-commercial uses of the work are permitted without any other permission from Dove Medical Press Limited, provided the work is properly attributable. To obtain permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.

Leave a Reply

Your email address will not be published.