Carta ao editor
Editor: We read with interest the article by Wang and You in the March 2013 issue of Radiology (1). They aimed to identify predictors of clinically important abnor- mal findings on computed tomographic (CT) images of the head among emer- gency department patients without a history of trauma.
The identification of clinical predictors of the ultimate utility of imaging examinations is an area of research with the potential to refine medical reasoning leading to further imaging examinations and to affect the efficient investment of available financial re-sources. Because there is a scarcity of studies examining the utility of head CT in patients without trauma, this study deserves special attention.
However, important limitations of the study must be noted. The explicit ones were pointed out by the authors: The retrospective design may have in-troduced bias to the data, and the het-erogeneity in CT requisitions (nonstan-dardized and nonuniversal but guided by previous physician’s assumptions in face of the physical examination find-ings) possibly changed pretest probabil-ity and also overestimated the predic-tors’ sensitivity and specificity.
We also need to draw attention to the omission of specificity and likeli-hood ratios. By analyzing the published data, notably those in Table 4 of their article, we can infer the following: (a) For the parameter “One or more of five clinical predictors, or age >70 y” (the independent predictors proposed in Table 3), the sensitivity was 96.0% but the specificity was 24.0%, resulting in a positive likelihood ratio of only 1.26 (95% confidence interval: 1.22, 1.31); and (b) for the parameter “One or more of five clinical predictors” or the one including “presentation with sei-zures,” the specificity and positive like-lihood ratio would be alike or even worse.
A positive likelihood ratio this low results in minimal or no change to the posttest probability (2–4). Considering the somewhat low prevalence of abnor-mal findings in this study sample, this is more worrisome. For example, applica-tion of the positive likelihood ratio of the proposed clinical predictors (1.26) to the pretest probability of 14.2% (the prevalence of abnormal CT findings in the article) would give us a similar post-test probability of 17.0% (5).
Identifying a set of predictors for abnormal findings in head CT images would require specificity to be as valued as sensitivity and the study design to preferably be prospective, including a universal CT requisition protocol, inde- pendent of a patient’s previous signs or symptoms.
Response
From Xi Wang, MD,* and John J. You, MD, MSc†‡ Departments of Radiology,* Medicine,† and Clinical Epidemiology & Biostatistics,‡ McMaster University, 1280 Main St West, Room HSC-2C8, Hamilton, ON, Canada L8S 4K1e-mail: jyou@mcmaster.ca
We are grateful for Dr Solla’s interest in and constructive comments about our work (1). We concur that the specific-ity and positive likelihood ratio derived from our data are low. However, as Dr Solla points out, clinical decision rules are valuable because of their ability to promote the more efficient use of re-sources. The principal opportunity for cost reduction therefore arises from the ability of a clinical decision rule to “rule out” the presence of a significant abnormality and thus enable physicians to safely avoid requesting a particular test, such as head CT. In other words, the utility of a clinical decision rule is largely determined by its sensitivity and negative likelihood ratio.
The negative likelihood ratio for pa-tients with one or more of the five clin-ical predictors identified in our study (ie, focal neurologic deficit, altered mental status, history of malignancy, nausea or vomiting, derangements in coagulation profile) or those older than 70 years is as follows: (1 – sensitivity)/specificity = (1 – 0.96)/0.24 = 0.17. If we apply a negative likelihood ratio of 0.17 to a population where the pretest probability of having a significant ab-normality at head CT is 14.3%, the posttest probability is reduced substan-tially to 2.8% (2). For the combination of “one or more of five clinical predic-tors, age >70 y, or presentation with seizures,” the associated negative likeli-hood ratio is 0.09, reducing the post-test probability further to 1.5%.
As stressed in our article, the clinical predictors of abnormal head CT identi-fied in our study require prospective val-idation before clinical application (3). Our findings represent the first step in the development of a clinical decision rule that has the potential to substan-tially reduce CT use in this patient popu-lation without missing clinically impor-tant neurologic abnormalities.
