Free
Correspondence  |   April 2009
When the Model Does Not Fit the Data, the Model Is Wrong
Author Affiliations & Notes
  • Dennis Fisher, M.D.
  • †P Less Than, San Francisco, California.
Article Information
Correspondence
Correspondence   |   April 2009
When the Model Does Not Fit the Data, the Model Is Wrong
Anesthesiology 4 2009, Vol.110, 947. doi:10.1097/ALN.0b013e31819b642c
Anesthesiology 4 2009, Vol.110, 947. doi:10.1097/ALN.0b013e31819b642c
To the Editor:—
Pühringer et al.  1 provide convincing evidence that sugammadex speeds the reversal of rocuronium and that the benefit increases with dose. Four figures display time to 90% train-of-four recovery as a function of sugammadex dose; in addition to raw data, each figure displays a curve based on fitting an exponential model to the data. Unfortunately, these curves provide misleading information about the dose-effect relationship. In that increasing the sugammadex dose should decrease (or not change) recovery time, the exponential model has the correct form (monotonic nonincreasing) but the wrong shape (as evidenced by its failure to match the data at low doses).1
The authors' intent in fitting this curve is to allow claims independent of the data. For example, based on their figure 1, one might claim that “following a dose of 2 mg/kg, time to 90% train-of-four is expected to be 30 min.” Yet, this claim is not supported by the data. Similarly, any claim about the expected response at doses not studied cannot be inferred from the curve drawn by the investigators.
The process of fitting the exponential curve is based on the authors' belief that they have a “model” for the relationship between dose and effect. Pharmacokinetic models are typically based on physiologic principles. For example, after bolus intravenous administration of a drug, samples drawn after the initial recirculatory phase typically show a monotonic decline that can be described by the sum of 1–3 exponentials; this model depends on the “reasonable” assumptions that clearance is constant and is proportional to plasma concentration. In contrast, there is no a priori  reason for the authors to assume that their dose-response relationship is described by an exponential equation.
The purpose of this communication is not to berate the authors. In Supplemental Materials (which few readers will examine), they acknowledge that “another nonlinear model [might] have better fitted the… results.” Instead, it is to remind future investigators (and reviewers and readers) of a maxim taught by our mentor, Lewis B. Sheiner, M.D. (1940–2004): “If the model does not fit the data, the model is wrong.”
†P Less Than, San Francisco, California.
Reference
Reference
Pühringer FK, Rex C, Sielenkämper AW, Claudius C, Larsen PB, Prins ME, Eikermann M, Khuenl-Brady KS: Reversal of profound, high-dose rocuronium-induced neuromuscular blockade by sugammadex at two different time points: An international, multicenter, randomized, dose-finding, safety assessor-blinded, phase II trial. Anesthesiology 2008; 109:188–97Pühringer, FK Rex, C Sielenkämper, AW Claudius, C Larsen, PB Prins, ME Eikermann, M Khuenl-Brady, KS