Criteria abstracted from The Users' Guide to Medical Literature, from the Health Information Research Unit and Clinical Epidemiology and Biostatistics, McMaster University

Highlighted lines and questions below provide links to the pertinent description of criteria in The EBM User's Guide, now available at the Canadian Centres for Health Evidence

Article Reviewed:

Please visit the new Evidence Based Journal Club Reviews

Intracranial hemorrhage in children younger than 3 years: prediction of intent.

Wells RG, Vetter C, Laud P.

Arch Pediatr Adolesc Med. 2002;156(3):252-7. [abstract; full-text for subscribers]

Reviewed by Geoff Capraro, MD and Scot Bateman, MD Boston Medical Center

Review posted August 26, 2002

I. What is being studied?

Study objective:

To determine, amongst children < 3 years with intracranial hemorrhage, whether certain head CT patterns can help discriminate intentional from unintentional injuries.

Study design

Retrospective, 10 year consecutive case series at a regional pediatric medical center

Inclusion criteria:

All head CTs over a 10 yr period reporting the presence of intracranial hemorrhage in children < 3 yrs old at one institution.

Exclusion criteria:

Children with hemorrhage because of prematurity, birth trauma, medical conditions, or intracranial surgical procedures.

Outcomes:

Sensitivity and specificity of CT imaging patterns for intentional head injury (which had an incidence of 50.5% in this population)

II. Are the results of the study valid?

Note: These questions follow from Randolph AG et al. Understanding articles describing clinical prediction tools. Crit Care Med 1998;26:1603-1612. [abstract]

1. Was a representative group of patients completely followed up? Was follow-up sufficiently long and complete? Was the sample representative of the larger population they were trying to sample from?

The 293 child sample included all infants and toddlers with intracranial hemorrhage diagnosed by head CT over a 10-year period at this regional medical center. The authors do not state the proportion of these patients that was referred versus primarily seen, which might matter for CT interpretation, nor, the average length of follow-up accomplished through their chart review. 148 (50.5%) patients were classified as intentional trauma (confession, or incompatible mechanism, or no explanation), 109 (37.2%) unintentional, and 36 (12.3%) intent uncertain. No information is given on the details of follow-up: how extensive or complete the reviewed medical records were, nor how frequent or accessible Òofficial child abuse investigationÓ reports were. It is not clear whether the latter reports were from child protection services, child protection team, and/or district attorneys. It is not clear whether more definitive determinations of intent might have been achieved. Given these ambiguities, follow-up was not complete.

2. Were all potential predictors included?

No. Predictors such as physical exam findings of patterned bruising or retinal hemorrhages were not included. The authors clearly stated which variables were considered in the prediction model, but not the rationale behind their selection.

3. Did the investigators test the independent contribution of each predictor variable?

Yes, a table is provided with univariate logistic regression results of odds ratios, confidence intervals, and p values, as well as a separate table containing the 4 significant variables determined by multiple regression analysis.

4. Were outcome variables clearly and objectively defined?

The authors are fairly clear and logical about the outcome variables defining intentional injury, but there is a great deal of subjectivity underlying in the details of these categorizations, so there may be significant mis-classification of patients along the main outcome measures.

There is no mention of the reliability or validity of the CT readings used as predictors. The authors reviewed the CTs blinded to the clinical history. Their categories were objective, based on anatomic regions for the most part (except brain edema). They state that the imaging categories were not mutually exclusive, however. They did not report whether there was more than one reviewer to test inter-rater reliability of their category assignments.

Again, these issues are critically important for determining whether one can confidently apply CT finding results to decision making. One needs to know that the categorization is reliable.

III. What are the results?

1. What is(are) the prediction tool(s)?

The authors attempted use the significant variables from the multivariate analysis to create a predictor tool, but the combinations of these variables entered in the model cannot be described in words or an equation. The authors found a model to predict intentional ICH with sensitivity of 84% and specificity of 83% given a 50% prevalence of intentional ICH. Ten combinations of their four significant variables (hygroma; convexity subdural hematoma without hygroma; absence of skull fracture; and inter-hemispheric subdural hematoma) were employed to predict intentional trauma.

If the authors had chosen to increase the specificity of the model, fewer combinations of CT findings might have been included, and this may be sensible, given that the patient population is already known to have intracranial hemorrhage, and their question is whether specific CT findings predict intentional injury. Indeed, stopping after the 6th combination of variables would have correctly classified 116 of 124 cases, for a specificity of 94%.

2. How well does the model categorize patients into different levels of risk?

Very poorly. The methodological shortcomings of assigning intent, together with a difficult-to-comprehend and perhaps, insufficiently specific model, leave the reader unclear on how well the model performs.

An example of specific troubles interpreting the results is found in the ability of hygroma to predict intent. The authors describe that this finding carries an odds ratio of 47.4, and that it was only found in patients with intentional injury. This may be quite misleading, since, although these findings were clearly on the initial CT, the authors provide no information on the timing of neuroimaging with respect to time of injury.

3. How confident are you you in the estimates of risk?

We are not, since the authors offer no information on the numbers of patients meeting intent classification by specific criteria. For instance, if all of the intentional injuries were classified as so because there was a confession, one would feel fairly confident about the author's classifications. Instead, they provide no specific numbers. The miss-classification of outcome (intent) makes assessment of predictors extremely difficult.

The authors do provide confidence intervals around their odds ratio estimates of risk, as well as the sensitivity and specificity of intent classification in the model, which respectively are: 84% (78-90%) and 83% (74-89%). These cutoff used to be included in this model is a probability of intentional injury of .45, roughly the same a s the entire cohort, so the clinical meaning of the combination variables included in the model is far from clear.

IV. Will the results help me in caring for my patients?

1. Does the tool maintain its prediction power in a new sample of patients?

Unknown. The model was not applied to a separate, validation sample. Using their prediction model on the 36 patients classified as intent uncertain, the authors predicted that 69%, or 25 additional patients suffered intentional trauma.

2. Are your patients similar to those patients used in deriving and validating the tool(s)?

The authors do not give a great deal of information on the demographics of their patients, but this probably does not matter a great deal. We do care for children younger than 3 years with intracranial hemorrhage, and could use better data about the meaning of specific findings.

3. Does the tool improve your clinical decisions?

Yes, but somewhat indirectly: in the text of the study, one reads that the majority (52%) of children with intracranial hemorrhage did not have a skull fracture, and that an even higher proportion (71%) of those with intentional injury had no fracture. These results will reinforce my practice of obtaining head CTs in the emergency department if there is an indication, and further restrict my use of skull films in cases of young children with suspected significant head trauma. This could translate into altered therapy when an occult, significant intracranial hemorrhage is identified by CT and requires hospitalization, ICU monitoring, or even neurosurgical intervention, but would have been missed by plain films.

The study alerts the reader to the high correlation of hygroma with intentional injury, and such a finding on head CT might prompt clinicians to more vigorously pursue a thorough abuse evaluation and safety planning for children related to the index patient.

4. Are the results useful for reassuring or counseling patients?

Probably not. As best as one can tell, there is no CT finding that can categorically determine intent.

 

Comments

Back to the EB Journal Club Index