Patients

This study made use of a database, as developed in a large ongoing diagnostic study in children with meningeal signs (9;10) (children from one month up to fifteen years of age) visiting the emergency department of the Sophia Children's Hospital, Rotterdam, The Netherlands, between 1988 and 1998. In this database, demographic data, signs, and symptoms at admission and data from hospitalization or outpatient follow-up were collected retrospectively from patient records. Data regarding diagnostic procedures such as laboratory tests of CSF, blood, stool, and urine specimens and radiographic tests at admission and in hospital, were retrieved from the computer-documented hospital information system. This system has been proved to be a reliable source for actual resource utilization of patients. Frequencies of the major resource use items were verified with data from the patient record.

Decision Rule

For extensive information on the derivation and validation of the decision rule, we refer to previous publications (9;10). In short, logistic regression analysis identified independent predictors for bacterial meningitis obtained from patient history, physical examination, and blood and CSF laboratory tests. By using the regression coefficients of the variables in the diagnostic models, the regression models were transformed into scoring algorithms.

Resource Use

Resources in the diagnostic course were emergency department visit, lumbar puncture procedure, laboratory tests (including hematology, blood chemistry, urinalysis, CSF cytology, and biochemical analysis and microbiology of blood and urine and CSF), and chest radiography. Resources in the treatment course included hospitalization (days in an intensive care unit, isolation, or pediatric ward), consultations by the pediatric specialist, laboratory and radiographic tests in addition to those performed in the diagnostic phase, and prescription of medication use either in hospital or during outpatient treatment (e.g., antibiotics, anti-epileptics, procedure of intravenous treatment). The actual resource use was based on the resource frequency in current practice as documented in the database. To estimate the expected resource use after applying the diagnostic decision rule, we simulated the consequences of all possible scenarios following the decision rule. Therefore, three groups of diagnoses were distinguished, according to their need of treatment: (i) patients with bacterial meningitis (BM), (ii) patients with serious bacterial infections other than meningitis (SBI; i.e., septicemia, pneumonia, urinary tract infection, or bacterial gastric-enteritis), and (iii) patients with viral or aseptic meningitis or with other self-limiting diseases (SLDs; i.e., upper respiratory tract infections, nonspecified viral infections, myalgia). Applying the decision rule resulted in the following possible scenarios: (i) patients with no indication for lumbar puncture, either ambulatory treated (SBI, n=7; SLD, n=118) or in hospital (SBI, n=4; SLD, n=4); (ii) patients with indication for lumbar puncture but no indication for empirical antibiotic treatment, either ambulatory treated (SBI, n=4; SLD, n=50) or in hospital (SBI, n=8; SLD, n=8); (iii) patients with indication for lumbar puncture and for empirical antibiotic treatment (BM, n=99; SBI, n=13; SLD, n=45).

In this estimation of patient numbers per scenario, we assumed not to miss any child with bacterial meningitis and to treat all patients correctly after applying the decision rule. All patients were assumed to have a similar risk of mortality and morbidity as in current practice. To make the cost estimation of the decision rule more realistic to practice, we made assumptions with regard to volume of diagnostic tests. In patients with BM or SBI, routinely applied laboratory and radiographic tests were assumed to be performed in the same frequency as they had been performed in the preexisting strategy (1). In SBI patients, the lumbar puncture may be saved by the diagnostic strategy if so predicted. In all patients with SLD, we assumed that costs of diagnostic tests other than in the prediction rule indeed were saved. However, because the distinction between SBI and SLD at the start of the evaluation will not always be clear, some additional diagnostic procedures (i.e., to exclude SBI) would be performed anyway in those with SLD. Hence, we assumed that these procedures would be performed in the same frequency in SLD patients as in SBI patients. Finally, in contrast with hospitalizations for empirical treatment in cases of suspicion of bacterial meningitis, we assumed that hospitalizations indicated by illness severity (i.e., hospitalized patients who did not actually undergo a lumbar puncture or who continued treatment when CSF cultures ruled out BM) were not affected by the rule.

Unit Costs

Average unit prices of diagnostic tests and treatment interventions were estimated in euros (€ 1=$0.946, January 2001), by using the internal unitary cost-estimation model of the Sophia Children's Hospital, Rotterdam, of 1996 (13). In this cost-estimation model, all hospital costs (including all general equipment, overhead, buildings, etc.) are exhaustively assigned to departments and their health care services. Unit costs of inpatient days and outpatient visits were directly derived from the cost-estimation model. For other health care services, additional data were collected about the estimated time the pediatrician and nurse spent per service. The time for an emergency department visit was estimated to be forty-five minutes for a pediatrician, forty-five minutes for a resident, and fifteen minutes for a nurse. Time for a lumbar puncture was estimated to be fifteen minutes for a pediatrician and forty-five minutes for a nurse. These estimates included the time for preparation of the patient and administration. Costs of radiographic tests were based on a separate costing study and included the time spent by the radiologist, materials, and radiographic equipment. Costs of medication were based on actual purchase prices plus an allowance for chemists' costs. Because costs of inpatient days were expected to be one of the major cost drivers, estimates of these costs were also obtained from a second hospital (Juliana Children's Hospital, The Hague; a large general pediatric hospital) and varied in a sensitivity analysis.

Analysis

Total costs were estimated in euros per patient by multiplying the unit costs per resource item with the diagnostic and therapeutic resource use as actually performed and as expected when applying the decision rule (given the above assumptions on resource use). In a cost-minimization analysis the possible cost savings due to the introduction of the diagnostic decision rule were compared with current practice. Application of the decision rule could reduce costs by avoiding (i) (unnecessary) diagnostic procedures, (ii) unnecessary administration of empirical antibiotics, and (iii) unnecessary or prolonged hospitalization. If additional tests were performed because of the decision rule, the costs of each scenario were increased with the costs of those procedures. In univariate sensitivity analysis, we assessed the impact of changes in our base-case input variables on the cost savings. Inpatient nursing costs were varied by 20%, which is a plausible variation of inpatient nursing costs between an academic hospital and a general hospital (8). The frequency of bacterial meningitis was reduced by 50%, with a proportional increase of the other two diagnostic categories (SBI and SLD). In a third sensitivity analysis, we assumed that all diagnostic tests and interventions other than indicated by the decision rule (e.g., for a differential diagnosis or for treatment purposes), indeed were saved in all patients. Therefore, in this analysis, we included costs of determining serum CRP and glucose, CSF cell count, glucose, and bacterial cultures, and costs of treatment as indicated by the rule only. Finally, we estimated the possible cost reduction by the rule by using data from a general pediatric hospital with a lower frequency of bacterial meningitis and lower inpatient nursing costs.