Clinical Features Suggestive of Meningitis in Children: A Systematic Review of Prospective Data

DISCUSSION

Information on efficient use of clinical findings is extremely important for clinicians. Useful features for estimation of probability of meningitis are those features that demonstrate the strongest LRs for presence or absence of disease. The LR of a clinical feature is the probability of that finding in patients with disease divided by the probability of the same feature in patients without disease (LRs range from 0 to infinity). Features with LRs equal to 1.00 have no diagnostic value, because it is equally likely to find the feature in those with the disease as in those without the disease. Features with LRs of >1.00 support the diagnosis of interest in magnitude of increasing numerical value. For features with LRs between 0 and 1.00, the smaller the LR, the less likely the disease.^18,19

Valuable features found in this review are listed in Table 3. On history, in order of decreasing magnitude, a caregivers' report of neck stiffness, bulging fontanel, seizures (outside the febrile-convulsion range), or reduced feeds raise concern about the presence of meningitis. On physical examination, in order of decreasing magnitude, the presence of jaundice, being toxic or moribund, or having meningeal signs, neck stiffness, bulging fontanel, Kernig sign, tone up, fever of >40°C, or Brudzinski sign all raise the probability of meningitis to varying degrees in the patient. Several other clinical features with LRs between 1.30 and 2.40, are less strong but warrant further study. Note that the sign petechiae is strong with an LR of 37.00 but was surprisingly only examined in a single small prospective study, and only 4 patients displayed the feature. Thus, relevance of this well-known feature is currently uncertain, and systematic prospective evaluations of it among large numbers of patients would provide clarity.

As an example of applicability, assuming statistical independence, a pretest probability of disease of 10%, and using the LR nomogram,²⁰ a combination of the presence of meningeal signs (LR: 4.50), a bulging fontanel (LR: 3.50), and a high fever (LR: 2.90) (thus, a combined LR of 45.60) raises an infant's probability of meningitis to 84%. Although the presence or absence of these findings, in combination or separately, hardly confirms or refutes a diagnosis of meningitis, they raise the probability high enough that an LP must be performed.

Each physician routinely incorporates a sense of the probability of disease through careful consideration of the clinical assessment, experience, and estimates of disease prevalence in the population. All of the studies included patients with a suspicion of meningitis or severe illness. The point prevalence of meningitis ranged from 4.2% to 19% across these studies; each prevalence reflects the clinical impression of possible meningitis (via initial inclusion in each study). The summary prevalence of these studies is 10%. This summary prevalence could be viewed as the posttest probability of the overall clinical examination, because all of the children were judged sick enough to undergo definite testing for meningitis. Assuming a prevalence of disease of 1%, the LR for the clinical impression of meningitis as its own independent “test” would be 11.00. Thus, the clinical suspicion of disease that a health care provider derives from clinical history and examination may, in itself, be a useful test that warrants follow-through to further diagnostic testing. However, although necessary for rapid comprehensive synthesis of complex clinical information, much is unknown about the process of clinical judgment and decision-making. Clinical impressions are prone to error, and efforts to minimize error by maximizing pretest probability through accurate clinical prediction or decision rules will offer improved patient care.^21,–,25

It seems clinically sensible that the combinations of some findings listed in Table 3 would have a greater impact on the probability of meningitis than the individual findings. Only 2 studies examined combinations of findings. It is unfortunate that original subject data from statistical models used in these studies were unavailable; thus, LRs could not be calculated. Nonetheless, Weber et al¹⁷ and Berkley et al^11,12 had constructed logistic regression models of varying combinations of features in an attempt to obtain sets of predictor variables with an optimal balance of sensitivities and specificities. The best combination model in the Weber et al study,¹⁷ which combined a history of seizures, being lethargic or unconscious, or having a stiff neck, had a sensitivity of 98% and specificity of 70%.¹⁷ This combination of features is a simplified Integrated Management of Childhood Illness referral criteria, a set of guidelines initially developed by the World Health Organization to identify sick children in need of referral.^17,26

However, Berkley et al¹¹ later tested this same model and found it to be only 85% sensitive and 59% specific. Further models from Berkley et al¹² included 1 with a high sensitivity of 97% but low specificity of 44% and combined nonmalarious fever with any 1 of the following: bulging fontanel; neck stiffness; cyanosis; seizures (outside of febrile seizure age range); partial seizures; and impaired consciousness. Another model combined impaired consciousness with any 1 of the following: bulging fontanel; neck stiffness; partial seizure; cyanosis; seizure (outside of febrile-seizure age range); it was found to be less sensitive (79%) but more specific (80%).¹² With a life-threatening highly morbid condition, diagnostic models that maximize sensitivity are essential. However, population overassessment, resulting from application of low-specificity models, is also of concern, particularly for regions in which distance or resource restrictions limit access to further care. Thus, the ideal clinical model for pediatric meningitis is still unclear, and prospective evaluation and validation of known and new prediction models in varying populations are imperative.

Although many of the symptoms and signs with available data demonstrated poor accuracy (Table 4), these findings have not been otherwise studied in combination. In addition, many other widely described features, otherwise reported in textbooks or review articles, have not been examined for validity in prospective studies. These commonly described clinical features warrant further prospective examination to confirm soundness of continued use in the context of meningitis.

When considering the results of this systematic review, clinicians should remain prudent regarding decision-making for young infants and particularly should not rely on the absence of archetypal features as reassurance of absence of disease. Several investigators from the included studies noted infants with meningitis who displayed few or no classic features of the disease. It is well accepted clinically that young infants with nonspecific yet concerning features such as fever, lethargy, poor feeding, or irritability, among others, must be approached with a high index of suspicion regardless of how well they appear, because the incidence of serious bacterial infection in this age group is much higher than that in older infants.