Hjort and Glad (1995) present a method for semiparametric density estimation. Relative to the ordinary kernel density estimator, this technique performs much better when a parametric vehicle distribution fits the data, and otherwise performs at broadly the same level. Jones, Linton, and Nielsen (1995) present a somewhat similar method for density estimation which has higher order bias for all sufficiently smooth densities. In this paper, we combine the two methods. We show that, theoretically, the desired properties of general higher order bias allied with even better performance for an appropriate vehicle model are achieved. Simulations suggest that the new estimator realises only a little of its theoretical potential in practice for small to moderately large sample sizes.

We present a rather thorough investigation of the use of kernel-based nonparametric estimators of the binary regression function in the case of a single covariate. We consider various versions of Nadaraya-Watson and local linear estimators. some involving a single bandwidth and others involving two bandwidths. The locally linear logistic estimator proves to be a good single-bandwidth estimator, although the basic Nadaraya-Watson estimator also fares quite well. Two-bandwidth methods show great potential when bandwidths are selected with knowledge of the target function, but much of their potential vanishes when data-based bandwidths are used. Likelihood cross-validation and plug-in approaches are the data-based bandwidth selection methods tested; both prove quite useful, with a preference for the latter. Adaptive two-bandwidth methods retain particularly good performance only in certain special situations (and separate estimation of the two bandwidths as for optimal density estimation is never recommended). We therefore propose a hybrid estimation procedure in which the local linear logistic estimator is used unless the ratio of (robust) variances of the covariate in the success and failure groups is greater than 2, in which case we switch to a two-bandwidth Nadaraya-Watson-type estimator, each using plug-in bandwidth selection.

BACKGROUND AND PURPOSE:This study describes the large variations in outcome after stroke between countries that participated in the International Stroke Trial and seeks to define whether they could be explained by variations in case mix or by other factors.METHODS:We analyzed data from the 15 116 patients recruited in Argentina, Australia, Italy, the Netherlands, Norway, Poland, Sweden, Switzerland, and the United Kingdom: We compared crude case fatality and the proportion of patients dead or dependent at 6 months; we used logistic regression to adjust for age, sex, atrial fibrillation, systolic blood pressure, level of consciousness, and number of neurological deficits. We used the frequency of prerandomization head CT scan and prescription of aspirin at discharge to indicate quality of care.RESULTS:The differences in outcome (all treatment groups combined) between the "best" and "worst" countries were very large for death (171 cases per 1000 patients) and for death or dependency (375 cases per 1000 patients). The differences were somewhat smaller after adjustment for case mix (160 and 311 cases per 1000 patients, respectively). Process of care may have accounted for some but not all of the residual variation in outcome.CONCLUSIONS:Adjustment for case mix explained only some of the variation in outcome between countries. The residual differences in outcome were too large to be explained by variations in care and most likely reflect differences in unmeasured baseline factors. These findings demonstrate the need to achieve balance of treatment and control within each country in multinational randomized controlled stroke trials and the need for caution in the interpretation of nonrandomized comparisons of outcome after stroke between countries.

OBJECTIVE:To determine how far the difference in published stroke case fatality between the Western General Hospital (WGH), Edinburgh and the Falkirk and District Royal Infirmary (FDRI) for the period 1990-93 can be explained by adjusting more fully for casemix.DESIGN:The cases were ascertained and followed prospectively at the WGH and retrospectively at the FDRI; casemix correction was performed using a validated logistic regression model.SETTING:The WGH is a teaching hospital and the FDRI a district general hospital.SUBJECTS:Four hundred and thirty seven patients with a verified acute stroke at the WGH; 471 patients assigned a cerebrovascular disease discharge diagnostic code at the FDRI.OUTCOME MEASURE:Thirty day case fatality.RESULTS:About half of the difference in the two hospitals' published stroke case fatality could be accounted for by variation in measured casemix. The residual difference in adjusted case fatality might have been due to differences in the structure of stroke care or simply to remaining differences in casemix. Full investigation of the cause was prevented by the destruction of the deceased patients records.CONCLUSIONS:Comparisons of routinely collected stroke outcomes will remain difficult to interpret unless casemix is properly accounted for and deceased patients' records stored for several years.

PURPOSE:The relative's questionnaire (RQ) was developed to assess outcome after brain injury. The present study investigated its test-retest reliability when used in a postal survey.METHOD:Hospital records were used to identify and contact 288 surviving patients treated for brain injury five to seven years earlier. Patients were sent a copy of the RQ (RQ1) and one month later a second copy (RQ2) was sent to those who returned RQ1.RESULT:Two hundred and eleven patients were successfully contacted, of whom 128 (61%) returned RQ1, and 94 of these (73%) returned RQ2. The reliability of items was variable, with most having a kappa value of > 0.6 suggesting 'substantial agreement' or better.CONCLUSION:The data presented suggest that the RQ is a reliable instrument in collecting outcome information in brain-injured patients by postal survey. Further research is recommended to test the suitability of the RQ for the use as a telephone interview.

To evaluate alternative methods of determining Glasgow Outcome Scale scores, a postal survey was made of 288 general practitioners and 128 relatives of patients who had sustained acute brain injuries 5-7 years previously. The Glasgow Outcome Scale score from the general practitioner and relative were compared with that calculated from questionnaire information by an experienced rater. There was poor agreement between general practitioner and rater (K = 0.17) and relative and rater (K = 0.35) scores. Both general practitioners and relatives indicated more favourable outcomes than the rater, with a higher level of agreement (K = 0.61) between them. When Glasgow Outcome Scale scores are used, the methods employed should be valid and reliable; failure to ensure this may be responsible for a considerable proportion of variability in reported studies of brain injury outcome.

The most common neuropathological substrates of dementia are Alzheimer’s disease, cerebrovascular disease, and dementia with Lewy bodies. A preliminary, retrospective postmortem analysis was performed of the relative burden of each pathology in 25 patients with predominantly Alzheimer’s disease-type dementia. Log linear modelling was used to assess the relations between ApoE genotype, Alzheimer’s disease, and cerebrovascular disease pathology scores. Sixteen of 18 cases (89%) with a Braak neuritic pathology score ⩽4 had, in addition, significant cerebrovascular disease, or dementia with Lewy bodies, or both. There was a significant inverse relation between cerebrovascular disease and Braak stage (p=0.015). The frequency of the ApoE-ε4 allele was 36.4%. No evidence was found for an association between possession of the ApoE-ε4 allele and any one pathological variable over another. In this series most brains from patients with dementia for which Alzheimer’s disease is the predominant neuropathological substrate also harboured significant cerebrovascular disease or dementia with Lewy bodies. The data suggest that these diseases are perhaps pathogenetically distinct, yet conspire to produce the dementing phenotype.

Objectives-To assess the prognostic value of summary measures of secondary physiological insult in addition to baseline clinical variables for patients with traumatic brain injury.Methods-A series of 110 patients with traumatic brain injury had data on intracranial pressure (ICP), arterial blood pressure (ABP), cerebral perfusion pressure (CPP), arterial O-2 saturation (SaO(2)), temperature in degrees C (Temp), and heart rate in beats/min (HRT) monitored and recorded every minute. Secondary insults were defined according to the Edinburgh University secondary insult grading system. The prognostic value of summary measures of these secondary insults was assessed by adding them to a prognostic model for survival at 1 year after controlling for baseline clinical variables using a previously validated model.Results-Of the eight secondary insults measured, only ICP added significantly to the prediction of survival in the first 72 hours after injury. The particular type of summary measure did not seem to influence the results. After the addition of ICP to the model, none of the other secondary insult measures could improve the predictive power of the model significantly.Conclusions-Early intracranial hypertension is confirmed as a sign of poor prognosis in patients with traumatic brain injury, even after controlling for baseline clinical variables. The value or otherwise of treating such secondary insults, however, can only be definitively established in the context of prospective randomised controlled trials. The specific pathophysiological evolution of secondary insults is still the subject of much research, and a clear understanding will be necessary before the development of specific treatments is feasible.

OBJECTIVES Prediction of patient outcome can be useful as an aid to clinical decision making, to explore possible biological mechanisms, and as part of the clinical audit process. Many studies have constructed predictive models for survival after traumatic brain injury, but these have often used expensive, time consuming, or highly specialised measurements. The aim of this study was to develop a simple easy to use model involving only variables which are rapidly and easily clinically achievable in routine practice. METHODS All consecutive patients admitted to a regional trauma centre with moderate or severe head injury were enrolled in the study. Basic demographic, injury, and CT characteristics were recorded. Patient survival at 1 year was used to construct a simple predictive model which was then validated on a very similar patient group. RESULTS 372 patients were included in the study, of whom 365 (98%) were followed up for survival at 1 year. Multiple logistic regression resulted in a model containing age (p<0.001), Glasgow coma scale score (p<0.001), injury severity score (p<0.001), pupil reactivity (p=0.004), and presence of haematoma on CT (p=0.004) as independently significant predictors of survival. The model was validated on an independent set of 520 patients, showing good discrimination and adequate calibration, but with a tendency to be pessimistic about very severely injured patients. It is presented as an easy to use nomogram. CONCLUSIONS All five variables have previously been shown to be related to survival. All variables in the model are clinically simple and easy to measure rapidly in a centre with access to 24 hour CT, resulting in a model that is both well validated and clinically useful.

Thrombolysis increases case fatality but reduces the proportion of disabled survivors in recent trials in acute ischaemic stroke, although some trials show much higher mortality rates than others. One possible explanation for the different outcomes between trials is that the treatment effect with thrombolysis varies with baseline prognostic factors such as stroke severity. We examined the interaction between baseline risk and thrombolysis on outcome using individual patient data from the Multicentre Acute Stroke Trial–Italy (MAST-I). A multiple logistic regression of the MAST-I data was performed to identify which factors, identifiable at randomisation, most strongly predict a poor functional outcome. We then stratified the patients into those with severe strokes and those with mild strokes and examined the effect of thrombolysis on (a) case fatality and (b) dependency at 6 months after the stroke in the 157 patients who received streptokinase alone and the 156 controls. Streptokinase was found to cause an absolute increase of about 3% in case fatality in both “severe” and “mild” strokes; however, there was a 12% reduction in the number of dead or dependent “mild” strokes but a 6% increase in “severe” strokes. The number of patients was small, and therefore neither finding was statistically significant. In this exploratory analysis, the hazard with streptokinase appears similar in “severe” and “mild” strokes, but the benefit may be greater in “mild” strokes. Thrombolysis may be more effective in patients with “mild” strokes, but more information is required to confirm this hypothesis.

Using an independent data set, the utility of the Glasgow Head Injury Outcome Prediction Program was investigated in terms of possible frequency of use and reliability of outcome prediction in patients with severe head injury, or haematoma requiring evacuation, or coma lasting 6 hours or more, in whom outcome had been reliably assessed at 6 to 24 months after injury. Predictions were calculated on admission, before evacuation of a haematoma, or 24 hours, 3 days, and 7 days after onset of coma lasting 6 hours or more. Three hundred and twenty four patients provided 426 predictions which were possible in 76%, 97%, 19%, 34%, and 53% of patients on admission, before operation, 24 hours, 3 days, and 7 days respectively. Major reasons for non-feasible predictions were that patients were paralysed/ventilated as part of resuscitation or management. Overall, 75.8% of predictions were correct, 14.6% were pessimistic (outcome better than predicted), and 9.6% optimistic (outcome worse than predicted). Of 197 patients (267 predictions) whose eventual outcome was good or moderate, 84.3% of predictions were correct. For death or vegetative survival (96 patients with 110 predictions), 83.6% of predictions were correct but for severe disability (31 patients with 49 predictions), only 12.2% were correctly predicted. The utility of the Glasgow Head Injury Outcome Prediction Program compares favourably with other outcome prediction algorithms for patients with head injury.

OBJECTIVE:Secondary intracranial hypertension has been linked to leukocytosis. We examined our data bank containing physiologic recordings and outcome data of severely head injured patients to investigate the relationship between delayed increases in intracranial pressure (ICP), defined as occurring after a 12-hr period of normal ICP values, and leukocytosis. DESIGN:A retrospective study of observational data. SETTING:Regional neurosurgical unit and intensive care unit. PATIENTS:Sixty-four patients suffered increased ICP >20 mm Hg. Thirty-five patients fulfilled selection criteria for delayed increases in ICP (group 1). Twenty-nine patients with increased ICP with no preceding or intervening periods of normal ICP were selected as a comparison group (group 2). MEASUREMENTS AND MAIN RESULTS:Comparison of 12-month outcome revealed that 11% of group 1 patients died, with 49% remaining severely disabled, in contrast to group 2, where 35% of patients died and 14% were left severely disabled (p = .021). The pattern of outcome was independent of monitoring time, or injury severity. Regression modeling was performed for prediction of delayed increase in ICP. Of 46 patients with an initial increase then decrease in leukocyte count in the first 48 hrs, 65% experienced delayed increases in ICP, as compared with 18% of the 11 patients without this pattern p = .01 1). CONCLUSIONS:Patients with delayed increases have a significantly different pattern of outcome. Change in leukocyte count from admission to day 2 is a significant predictor of such a delayed increase.

The strain-gauge Codman MicroSensor intracranial pressure (ICP) transducer has shown consistently good laboratory performance. To assess the practical performance of the system in patients following acute brain injury, 10 patients were fitted with a MicroSensor and a second ICP monitor. In five cases this was a fibre-optic transducer and in five cases an intraventricular fluid-filled device. Paired ICP values were recorded every 5 min. ICP values ranged from 0 to 31 mmHg. Altman-Bland plots showed that individual readings could differ by as much as 9 mmHg. Further analysis showed that much of this disagreement could be explained by a constant offset on each occasion. Comparison traces of ICP in individual patients show high agreement in timing and size of changes. The unexplained constant offset leads to uncertainty about the true ICP. Treatment decisions are often based upon absolute levels of ICP and patient care may therefore differ depending upon the monitor used.

We propose and investigate two new methods for achieving less bias in non‐ parametric regression. We show that the new methods have bias of order h4, where h is a smoothing parameter, in contrast to the basic kernel estimator’s order h2. The methods are conceptually very simple. At the first stage, perform an ordinary non‐parametric regression on {xi, Yi} to obtain m^(xi) (we use local linear fitting). In the first method, at the second stage, repeat the non‐parametric regression but on the transformed dataset {m^(xi, Yi)}, taking the estimator at x to be this second stage estimator at m^(x). In the second, and more appealing, method, again perform non‐parametric regression on {m^(xi, Yi)}, but this time make the kernel weights depend on the original x scale rather than using the m^(x) scale. We concentrate more of our effort in this paper on the latter because of its advantages over the former. Our emphasis is largely theoretical, but we also show that the latter method has practical potential through some simulated examples.

We consider many kernel-based density estimators, all theoretically improving bias from O(h 2), as the smoothing parameter h → 0, to O(h 4). Examples include higher-order kernels, variable kernel methods, and transformation and multiplicative bias-correction approaches. We stress the similarities between what appear to be disparate approaches. In particular, we show how the mean squared errors of all methods have the same form. Our main practical contribution is a comparative simulation study that isolates the most promising approaches. It remains debatable, however, as to whether even the best methods give worthwhile improvements, at least for small-to-moderate sample exploratory purposes.

OBJECTIVES:The use of multimodality monitoring of patients in the intensive care unit (ICU) and the subsequent collection and analysis of such data are increasing. The aim of this work was to assess the importance of recording complementary textual data referring to patient care maneuvers, calibrations, and other incidents, in addition to the raw numerical values. DESIGN:A retrospective analysis of multimodality monitoring data, which included comments entered concurrently at the bedside, collected from head-injured patients admitted to an ICU. PATIENTS:One hundred forty-seven patients with a postresuscitation Glasgow Coma Scale score of < or = 12 were monitored for a total of nearly 1 million minutes on up to eight commonly used channels. MEASUREMENTS AND MAIN RESULTS:Approximately 13,000 comments were added to the raw data at the time of collection. The data were subsequently validated using these comments as indicators of artifactual values. The comments were classified into a surprisingly small number of important categories, with the most frequent referring to monitor calibrations and regular ICU care maneuvers. The difference between validated and unvalidated data on the quantity of secondary insult observed was in some cases nearly 50%. CONCLUSIONS:This work demonstrates that such textual information should be recorded concurrently with the raw monitoring values to ensure proper interpretation of the data in any retrospective analysis. Furthermore, it also suggests that a small number of prespecified categories could be used in the on-line validation of such data.

The Glasgow Outcome Scale (GOS) is the most widely used outcome measure in head injury research. However, it is a global and relatively insensitive measure, precluding any description of the types of impairments that lead to the disability. The Edinburgh Extended Glasgow Outcome Scale (EEGOS) was devised as a new outcome measure that retains the advantages of the existing GOS but allows comparison of patterns of recovery in different areas of function; behavioural, cognitive and physical. This report describes pilot studies of the EEGOS used retrospectively, and in 'live' face-to-face interviews. The results show raw percentage agreements of 45%, 60% and 70% in the retrospective study, and 83%, 78% and 83% in the 'live' study. These results demonstrate that the inter-rater reliability of the EEGOS is comparable to that of the GOS applied in similar situations.

After acute brain injury there may be increased intracranial production of cytokines, with activation of inflammatory cascades. We have sought to determine if a transcranial cytokine gradient was demonstrable in paired sera of 32 patients requiring intensive care after acute brain injury. The difference between concentrations of IL-1 beta, IL-6, IL-8 and TNF alpha in jugular venous and arterial serum was measured on admission, and at 24, 48 and 96 h after the primary injury. There were no differences in IL-1 beta, IL-8 or TNF alpha, but median gradients of 6.7 and 11.5 pg ml-1 for IL-6 were demonstrated in the traumatic brain injury (n = 22) and subarachnoid haemorrhage (n = 10) groups, respectively (normal values in serum < 4.7 pg ml-1; P < 0.001 both groups). This suggests that there is significant production of IL-6 by intracranial cells after acute brain injury. Therapy directed towards combatting the negative effects of IL-6 may potentially benefit patients who have sustained an acute brain injury.