Variation in Severity-Adjusted Resource use and Outcome for Neurosurgical Emergencies in the Intensive Care Unit

We used admissions between 2015 and 2017 from ICUs with cost data available [4]. We included patients with emergency admissions to ICUs with full in-house neurosurgical service (six university hospital ICUs: four in Finland and one each in Estonia and Switzerland). We considered as neurosurgical emergencies patients admitted for a nontraumatic ICH, SAH, isolated TBI, and multitrauma TBI. We excluded transfers from other ICUs; nonemergency admissions; those with an unknown hospital discharge status; readmissions; those with a missing Therapeutic Intervention Scoring System (TISS)-76 score, missing sex, or missing Glasgow Coma Scale (GCS) score; and those admitted as potential organ donors (Fig. 1).

We assessed resource use using ICU length of stay (LOS), daily collected extended TISS scores (TISS-76 scores, including 17 additional items (Supplementary Information eTable 1), referred to as “TISS”), and direct ICU costs (salaries, drugs, fluids, disposables) as previously described [4]. To include physicians’ salaries allocated to other budgets or simultaneously covering other services, physician staff organization, rotations, and in-house and on-call coverage were clarified with each ICU leader, and costs were allocated based on consensus time estimates [4].

Calculation of costSRUR has been described in detail, including a practical example [4]. Briefly, all yearly ICU admissions were stratified according to Simplified Acute Physiology Score II (SAPS II) scores (0–9, 10–19,…, 80–89, > 90) [7]. The expected resource use per survivor in each SAPS II stratum was the sum of LOS days or TISS scores divided by the number of hospital survivors. For each ICU, the number of survivors multiplied by the expected resource use per survivor in each stratum was calculated, and the sum of all strata was the expected resource use for each ICU.

In the present study, the observed and expected resource uses were calculated specifically for the subgroups of interest (nontraumatic ICH, SAH, isolated TBI and multitrauma TBI, and nonneurosurgical emergencies).

First, the proportion of all ICU direct costs was assigned to the study cohort in each ICU as equal to the study cohort’s proportion of the ICU’s total LOS and TISS, respectively. The sum of these costs in all ICUs provided the total direct costs of the study cohort.

Second, we applied in the diagnostic subgroups the same procedure as described above to calculate the expected resource use per survivor after assigning the proportion of total resource use (as direct costs, LOS, and TISS) to each diagnostic subgroup: the expected costs to produce a hospital survivor in each SAPS II stratum were calculated as (expected resource use LOS/TISS per survivor) × (mean cost of LOS/TISS). For each ICU, the sum of expected direct costs in all strata (number of survivors × expected costs to produce a hospital survivor) was the expected total direct cost [4].

The cost-based SRURs for each ICU were then calculated as observed/expected total direct costs based on LOS (costSRUR_LOS) and direct costs based on TISS (costSRUR_TISS).

We used a fixed exchange rate of 1.09 Swiss franc to 1.00 euro (EUR), without inflation adjustment, and we adjusted the costs for purchasing power parity [4, 8].

We calculated SMR for all ICUs as observed/predicted hospital mortality. We calculated the predicted hospital mortality using a recent prediction model customized for FICC [9]. Briefly, the risk model includes patient age, a modified SAPS II score (without age and admission type), patient premorbid functional status, and APACHE III diagnosis and admission type (emergency for all). GCS score was defined according to the SAPS II criteria as the worst score during the first 24 h in the ICU or the last score preceding sedation for intubated/sedated patients.

We followed a similar analysis strategy as previously described [4]. We described the study population by frequencies (n), percentages (%), and medians and interquartile ranges. We used box plots to describe SMR and costSRUR measures across the different diagnostic groups. We used gamma distributed regression models to investigate a priori defined ICU-related factors (total number of ICU beds, full-time equivalent [FTE] physicians-to-bed ratio, organization type [independent organization vs. part of another department], median SAPS and SMR) associated with costSRUR_LOS and costSRUR_TISS. We investigated the ICU-related factors' association with SMR and costSRURs on the ICU level for each diagnosis group separately. To avoid overfitting of the model, we reported only the bivariable results [10].

In the primary analysis, we excluded patients admitted as potential organ donors from both resources and outcomes [9]. We conducted three sensitivity analyses in which we (1) included used resources but excluded the number of survivors for patients admitted as potential organ donors and (2) included used resources and the number of survivors for patients admitted as potential organ donors.

Continuous variables were standardized (centered and expressed per 1 standard deviation increase) and relative risk estimates reported with 95% confidence intervals. All analyses were performed in R version 4.1.2 (R Team Core. R: A language and environment for statistical computing; R Foundation for Statistical Computing, Vienna, Austria).

Of all 28,674 patients treated in a university hospital with in-house neurosurgical services not fulfilling one of the other exclusion criteria (eTable 2), 311 were admitted as potential organ donors (Fig. 1). Of the 28,363 remaining patients, 6,162 (22%) were admitted with a neurosurgical disease. Patients with a neurosurgical diagnosis had lower GCS scores, had lower SAPS II scores, were more frequently admitted after surgery, more often had a normal premorbid functional status, and had higher extended TISS scores, longer ICU stays, and a lower hospital mortality rate (9.3% vs. 13.0%) than patients with a nonneurosurgical diagnosis (Table 1). Of the neurosurgical diagnoses, nontraumatic ICH was the most common diagnosis (41% of all admissions), followed by isolated TBI (24%), SAH (23%), and multitrauma TBI (13%).

Patients with multitrauma TBI were the youngest patient group, and patients with nontraumatic ICH were the oldest patient group. Women were overrepresented in the SAH group. SAPS II score distribution was similar between the diagnostic groups. Patients with SAH and multitrauma TBI had higher TISS scores and longer LOSs than the other groups (Table 2).

There were some differences in patient characteristics between units (eTable 3). Median age varied between 59 and 67 years, median GCS score between 5 and 13, and normal preadmission premorbid functional status between 56 and 94%. The proportion of neurosurgical-to-total admissions varied between 19 and 23%. Hospital mortality increased with a decreasing GCS score, with some notable differences in-hospital mortality rate among patients with GCS ≤ 8 across units (eFig. 1).

A comparison between SMR, costSRUR_TISS, and costSRUR_LOS is shown in Fig. 2 and eTable 4. We found no significant differences in SMR or costSRUR measures between the neurosurgical and nonneurosurgical diagnoses or within the neurosurgical diagnoses.

The annual direct costs for all ICU emergency admissions varied from 67.1 million EUR to 70.4 million EUR (TISS-based cost separation) and from 70.0 million EUR to 72.9 million EUR (LOS-based cost separation). The corresponding direct costs for neurosurgical emergency admissions represented 23.6–23.7% (TISS-based) and 25.7–26.0% of direct costs for all emergency admissions.

The mean costs per patient with neurosurgical diagnoses were highest for patients with SAH (12,466 EUR) and patients with multitrauma TBI (11,115 EUR) and lowest for patients with isolated TBI (6,978 EUR). The mean cost per patient with other diagnoses was 6,749 EUR (eTable 5). The mean costs per LOS day did not notably differ across groups (1,981–1,988 EUR per LOS day, whereas the costs per TISS score were higher in patients with neurosurgical diagnoses (50.6–54.5 EUR per TISS score vs. 48.2 EUR per TISS score in patients with other diagnoses).

The direct costs per survivor increased exponentially with increasing SAPS II category for all diagnoses, except for patients with SAH, in which the cost per survivor in the highest SAPS II categories, containing 1% of admissions, decreased (eTable 6, eFig. 2).

We found an association between higher costSRUR_TISS and costSRUR_LOS and lower SMR when combining all neurosurgical diagnoses (Fig. 3, eTable 7). Separating the neurosurgical diagnoses, the association between higher costSRUR_TISS and costSRUR_LOS and lower SMR was apparent only for patients with nontraumatic ICH.

A higher FTE physician-to-bed ratio was associated with higher costSRUR_LOS and costSRUR_TISS in patients with SAH (Fig. 4, eTable 8). A higher FTE physician-to-bed ratio was associated with a lower SMR for the nonneurosurgical diagnoses but not for any of the neurosurgical diagnoses.

An independent organization type was associated with lower costSRURs (costSRUR_LOS, costSRUR_TISS) in patients with nontraumatic ICH, isolated TBI, and multitrauma TBI. For patients with nontraumatic ICH, an independent organization type was associated with a higher SMR in the primary and both sensitivity analyses. Furthermore, a higher SMR was associated with a lower costSRUR_LOS in patients with nontraumatic ICH.

There was an association between a higher total number of beds and a higher SMR in patients with nontraumatic ICH and isolated TBI. The total number of beds was not associated with either of the SRUR measures.

A higher median SAPS II was associated with lower costSRURs in patients with SAH.

All bivariable associations remained statistically significant in both sensitivity analyses, with the exception of the association between organization type and lower costSRUR_TISS for patients with nontraumatic ICH and the association between a higher FTE physician-to-bed ratio and costSRUR_TISS for patients with SAH in the sensitivity analysis in which resources used for potential organ donors were included, but the number of survivors of the potential organ donors were excluded.

There are some limitations that should be acknowledged. First, we used a SAPS II–based case-mix model [9] that may not be optimal for neurosurgical diseases. It is possible that SAPS II overestimates the injury severity in neurosurgical patients who may have an initially lowered but correctable GCS score. For example, up to 50% of patients with SAH develop acute hydrocephalus requiring external ventricular drainage, which can temporarily lower the noted GCS score [17], or a patient with an unknown GCS score before starting sedation was given a GCS score reflecting sedation. However, because the FICC database does not contain data regarding EVD placement or the exact time-point of GCS assessment, we were unable to control for this. Second, due to the limited number of patients in the separate groups, we used bivariable modeling, which may not capture interactions between the variables. Third, most of the confidence intervals that do not include 1.0 are very close to it, and their relevance needs to be interpreted cautiously. Fourth, intercountry variations in organ donation policies may affect our results. Among the included three countries, the number of organ donations is the highest among Finnish centers, being in 2021 on average 22 of 1,000,000 compared with 16 of 1,000,000 in Estonia and 19 of 1,000,000 in Switzerland [18]. Because most of the included patients come from Finnish centers, this might possibly affect our results. Importantly, patients admitted as potential organ donors have an expected mortality of 100%, which is, in many instances, higher than the expected mortality according to their SAPS II score. Still, the sensitivity analyses did not alter our results, and thus the impact on variations in organ donations is probably negligible. Fifth, although it is an important measure for ICU performance, hospital mortality may not adequately reflect outcomes after ICU-treated neurosurgical emergencies, in which long-term neurological outcomes are of more relevance. For example, in a previous study, it was shown that patients with nontraumatic ICH had the poorest prognosis and highest costs per independent survivor out of several neurosurgical emergencies [19].