Ned Glick

Additive estimators for probabilities of correct classification

Abstract Several methods for estimating a sample-based discriminants probability of correct classification are compared with respect to bias, variance, robustness, and computation cost. “Smooth” modification of the counting estimator, or sample success proportion, is recommended to reduce bias and variance while retaining robustness. Also the “bootstrap” method of Efron (8) can approximately correct an additive estimators bias using an ancillary computer simulation. In contrast, bias reduction achieved by the popular “leave-one-out” modification of counting method is vitiated by corresponding increase in variance.

Sample-Based Classification Procedures Derived from Density Estimators

Abstract Parametric or nonparametric density estimators from a mixture of K distributions can be used to estimate the non-error rate of an arbitrary classification rule—and to construct a rule which maximizes estimated probability of correct classification. (For two multivariate normal distributions with common covariance matrix, this general criterion yields the usual linear discriminant.) Such a sample-based rule is asymptotically optimal under very general conditions. Often its “apparent” non-error rate exceeds its true rate and is even optimistically biased as an estimator of the (unknown) optimal rules non-error rate; but the apparent rate converges to this optimum.

A research tool for measurement of recovery from sedation: The vancouver sedative recovery scale

The need for a research tool to measure recovery from sedation was identified during the design phase of a study investigating sedative protocols following open heart surgery in children. A thorough review of the literature failed to show any scales that measure degree of sedation in children at various times after initial awakening. The Vancouver Sedative Recovery Scale (VSRS) was developed through an iterative process during which we identified numerous indicators of levels of alertness among sedated children, and then determined the applicability and face validity of these indicators. The VSRS evaluated in this study consists of 12 distinct items that encompass three categories of indicators (response; eye appearance and function; and body movement). Total possible VSRS scoring ranges from 0 to 22 (higher score indicating more alert) because some of the 12 items have more than two rating levels. The VSRS was administered to 82 pediatric intensive care unit and postanesthesia recovery patients, with each patient assessed simultaneously by at least two observers. Internal consistency as measured by Cronbachs alpha was excellent: 0.85. Interobserver agreement or reliability as measured by intraclass correlation was also very high: 0.90; and for individual items Cohens kappa ranged from 0.65 to 0.89. We consider the VSRS to be a good beginning in our effort to quantify level of alertness after sedation in the pediatric patient population.

Randomized Trial of a Health Promotion Program For Frail Elders

This study evaluates a local health promotion project that may be widely adaptable to assist frail elderly persons to live longer at home. Subjects, enrolled in New Westminster, B.C., were men and women aged 65 and over living in their own homes but assessed and newly admitted to “personal care at home” by the Long Term Care (LTC) program of the B.C. Ministry of Health. About 90 per cent of eligible clients consented to participate. Randomized to Treatment or Control, they were followed for three years. Controls ( n = 86) received standard LTC services, which included screening and pre-admission assessment, arrangement/purchase of needed services and review at three months and at least yearly thereafter. The Treatment group ( n = 81) received standard LTC services plus visits from the project nurse who helped each subject to devise a personal health plan based on his or her needs in the areas of health care, substance use, exercise, nutrition, stress management, emotional functioning, social support and participation, housing, finances and transportation. The visits concentrated on setting goals and developing personal health skills, with referral to appropriate community services. An additional group of LTC clients ( n = 81) from the adjacent community of Coquitlam was also followed. Success or “survival” was defined as “alive and still assessed for care at home”. After three years the “survival rate” for the Treatment group was 75.3 per cent, compared with 59.3 per cent for the Control group and 58.0 per cent for the Coquitlam group. Standard Kaplan-Meier “survival” graphs show that Treatment subjects were more likely to be alive and living at home at every time point during the three years. Differences between the Treatment and Control groups were statistically significant ( p ≤ 0.05) both for simple cross-tabulations of care status at 24 and 36 months and in tests comparing “survival” curves. The results are especially striking because Control subjects received LTC services in a geographic area that offers universal access to health care and community resources and because the Control data were concurrent, not historical.

Separation and probability of correct classification among two or more distributions

A function on theK-fold product of a set in normed vector space will be called a separation measurement if, for any collection ofK points, the function is bounded below and above, respectively, by maximum and total distance between pairs of points in the collection. Separation measurements are relavent toK-sample hypothesis testing and also to discrimination amongK classes, and several examples are given. In particular, ordinaryL1 distance between integrable functions can be generalized to a non-pairwise separation measurement for densitiesf1,f2,...,fK inL1[μ]; and this separation is a linear transform of the optimal discriminants probability of correct classification.

A comparison of regularly dosed oral morphine and on-demand intramuscular morphine in the treatment of postsurgical pain

A randomized, placebo-controlled, double-blind clinical trial was conducted to compare the use of regularly dosed po morphine and on-demand im morphine in 47 patients undergoing total hip arthroplasty. Patients were randomized to receive either 20 mg (initial dose) of regularly dosed morphine (every four hours po) plus breakthrough pain medication on-demand consisting of both 10 mg morphine po and placebo im, or an equivalent regularly dosed oral placebo (every four hours) with breakthrough pain medication consisting of oral placebo and 5–10 mg morphine im. Subsequent to each request for break-through pain medication, the next regularly dosed oral solution was increased by 5 mg (or equivalent volume of placebo) to a maximum of 40 mg po Q4H. Time-averaged pain scores were lower on both postoperative day 1 and 2 in the group receiving regularly dosed morphine po (P < 0.05). Fewer patients requested breakthrough pain medication on both days in the oral morphine group. The incidences of nausea and vomiting, and of decreased respiratory rates were similar in both groups. Regularly dosed oral morphine is inexpensive and should be compared to other methods of opioid delivery.RésuméUne étude randomisée, contrôlée avec placébo et à double insu est réalisée dans le but de comparer l’utilisation de morphine po avec la morphine sur demande im chez 47 patients après une arthroplastie totale de la hanche. La randomisation est effectuées de façon à ce que les patient reçoivent: soit morphine 20 mg po (dose initiale) répétée régulièrement (aux quatre heures) avec l’addition, pour les percées douloureuses sur demande de morphine 10 mg po ou d’un placébo im; soit un placébo oral administré avec la même régularité (aux quatre heures) avec l’addition pour les percées douloureuses de placébo po et morphine 5–10 mg im. Subsequemment à chaque demande supplémentaires pour des percées douloureuses, la dose régulière suivante est augmentée de 5 mg (ou d’un volume équivalent de placébo) jusqu’à un maximum de 40 mg po aux quatres heures. Les scores d’évaluation moyens sont plus bas pour les deux premiers jours postopératoires pour le groupe qui regoit de la morphine régulièrement (P < 0,05). Moins de patients reçoivent de la morphine pour des percées douloureuses pendant ces deux jours dans le groupe morphine po. L’incidence des nausées et vomissements et de bradypnée est la même dans les deux groupes. La morphine administrée po régulièrement coûte peu et devrait être comparée aux autres méthodes d’administration.

The Vancouver sedative recovery scale for children: validation and reliability of scoring based on videotaped instruction

We originally developed and tested the Vancouver Sedative Recovery Scale (VSRS) to measure recovery from sedation following paediatric open heart surgery and reported excellent clinical inter-observer reliability. We now report a new study using videotape instruction of novice raters and videotaped case examples to determine whether the instruction produces adequate skill with the VSRS. Inter-rater reliability was assessed using videotapes of 16 children across a range of ages (six months to six years), and all levels of sedation (unresponsive to fully awake). Variably randomized subsets of six of the 16 test cases were randomly assigned to be rated by each of 16 video-instructed ICU staff volunteers, according to a balanced incomplete block design, such that every pair of raters assessed two children in common. The validity of the ratings from the video-instructed raters was assessed by comparison with “gold standard” scores from two experts who rated all 16 children clinically as the test cases were videotaped. The experts were in agreement themselves (intraclass correlation of 0.976). The correlation between the novice scores (average of six ratings per video) and the live clinical scores (average of two expert ratings) was 0.977 over the 16 test cases. On average, the mean expert rating was slightly higher, but the difference was negligible. (The differences between the mean ratings of the experts and novices for 13 of the 16 videos are very close to zero, while the other three differences, from technically less good videos, are two standard deviations away from zero). The VSRS, video instruction method and incomplete block design may be of use to other clinicians and investigators.RésuméPour mesurer le réveil et la sédation après la chirurgie à coeur ouvert chez l’enfant, nous avons développé et vérifié l’échelle de réveil et de sédation de Vancouver (VSRS) dont nous avons déjà communiqué la fiabilité. Nous pulions maintenant une nouvelle étude au cours de laquelle nous utilisons une vidéocassette pour former les appréciateurs et des exemples sur vidéocassette pour déterminer si cette méthode suffit pour produire une compétence suffisante avec le VSRS. La variabilité entre appréciateurs est évaluée avec des vidéocassettes de 16 enfants d’âges différents (de six mois à six ans) et à tous les degrés de sédation (de l’insensibilité au réveil complet). Des souscatégories variables aléatoires de six des 16 observations expérimentales sont attribuées au hasard pour évaluation à chacun des 16 préposés volontaires déjà initiés à la vidéo à l’unité des soins intensifs, suivant un modèle de bloc équilibré incomplet de sorte que chaque paire d’appréciateurs doit évaluer deux enfants en commun. La validité des évaluations des appréciateurs formés par vidéo est établie par comparaison avec les scores de deux experts qui ont classifé par la clinique les 16 enfants au moment où on les enregistrait sur vidéo. Les experts sont d’accord entre eux (corrélation 0,976). La corrélation entre les appréciateurs novices (en moyenne six évaluationspar vidéo) et les scores de références mesurés sur le vif (moyenne de deux évaluations d’experts) est de 0,977 pour les 16 observations. Les évaluations moyennes des experts sont légèrement plus élevées, mais cette différence est négligeable. Pour 13 des 16 vidéos, la différence entre les évaluations des experts et des novices est très près de zéro, alors que les trois autres différences pour des vidéos de moins bonne qualité technique s’éloignent de deux écarts-types du zéro. Le VSRS, la formation par vidéo et le modèle de bloc équilibré incomplet peuvent être utiles à d’autres cliniciens et chercheurs.

Midazolam following open heart surgery in children: haemodynamic effects of a loading dose.

Our objective was to establish the safety and effectiveness of a loading dose of midazolam for postoperative sedation of children recovering from open heart surgery; a prospective randomized placebo‐controlled double‐blind study was done with subjects randomized to three groups according to loading dose. I=0.08 mg·kg−1; II=0.04 mg·kg−1; and III=0.00 mg·kg−1 (placebo). An open label continuous midazolam infusion protocol followed. Haemodynamic parameters were monitored. The study was discontinued following an adverse event involving the 23rd subject. When data for all 23 subjects were combined, there was a mean decrease of 10% in blood pressure (BP) 30 min after the loading dose (P<0.001). Heart rate change was less significant. Clinicians identified four hypotensive episodes as temporally associated with the midazolam load, two each in Groups I (0.08 mg·kg−1) and III (placebo). One subject in Group I (the 23rd) became hypotensive within five min of receiving the loading dose, had a difficult clinical course and died four weeks postoperatively. We cannot conclude that the loading dose of midazolam had any systematic haemodynamic effect in our study population. Although the clinical course of the 23rd subject suggests a subset of more susceptible children (those who receive opioid analgesia with midazolam, are volume‐restricted, and/or undergo more complex forms of surgical correction), many critical care patients are inherently physiologically unstable, and concluding clinically that blood pressure fluctuation is drug related may be erroneous.

Apical root strain as a function of post extension into a composite resin core

This study tested the hypothesis that there is no difference in the amount of strain measured in the root of an extracted tooth at the apical end of a prefabricated post, whether a composite resin core completely covers the head of the post or the post is exposed at the occlusal surface of the core. Twenty extracted teeth were measured (1) after endodontic treatment (baseline), (2) after a composite resin core was fabricated over a prefabricated post, and (3) after a cast crown was cemented. The teeth were randomly divided into two groups after baseline testing. In group 1, the post was buried within the core, and in group 2, the head of the post was exposed on the occlusal surface of the core. Analysis of results indicated a statistically significant decrease in strain from the baseline to the crown when 1 mm of composite resin covered the head of the post. However, based on the small strain values measured, the difference may not be clinically significant.