Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Statistical Power”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,135 records · Page 63Linked to original sources
PubMedDate not supplied

Mega-trials: methodological issues and clinical implications.

A recent development of the therapeutic trial has been the mega-trial: a large, simple randomised trial analysed on an 'intention to treat' basis. Mega-trials have advantages in terms of increased statistical power, but also raise several new questions of interpretation. In mega-trials, randomisation serves to achieve identical allocation groups in a situation where there is poor experimental control and a large measure of between-subject variation. The results of mega-trials cannot readily be generalised because their conclusions are observations, not casual hypotheses, and are therefore not testable. In this sense, mega-trials can be repeated but cannot be replicated. Basic science and clinical science both seek understanding at the level of the individual subject; but in a mega-trial, analysis is only meaningful at the group level. The non-scientific nature of mega-trials derives from their methodology, which dispenses with the scientific aim of maximum experimental control to remove or minimise bias, and instead uses randomisation to achieve an equal distribution of bias between groups.

Bias↗
PubMed2004

Survival analysis of dropout patterns in dieting clinical trials.

Subjects who withdraw from diet clinical trials are a drain on limited resources and reduce statistical power. Dropout pattern data, collected during a clinical trial for which the primary findings compared weight loss from three dieting protocols, are examined using survival analysis and found to be exponentially distributed. The predicted probability of remaining in the study is 83% for 30 days and 60% for 84 days. Survival analysis methods consider subjects who did not return after the initial visit and others who may have continued dieting beyond study termination. When applied to clinical trials, this type of analysis provides valuable information for planning and budgeting of future trials. Inclusion of a 1- to 2-week run-in period at the beginning of the study may improve retention. Otherwise, the diet researcher should consider increasing initial randomized sample size by approximately 10% to 25% as an allowance for early withdrawals.

Body Mass Index↗
PubMed2006

Adjusting batch effects in microarray expression data using empirical Bayes methods.

Non-biological experimental variation or "batch effects" are commonly observed across multiple batches of microarray experiments, often rendering the task of combining data from these batches difficult. The ability to combine microarray data sets is advantageous to researchers to increase statistical power to detect biological phenomena from studies where logistical considerations restrict sample size or in studies that require the sequential hybridization of arrays. In general, it is inappropriate to combine data sets without adjusting for batch effects. Methods have been proposed to filter batch effects from data, but these are often complicated and require large batch sizes ( > 25) to implement. Because the majority of microarray studies are conducted using much smaller sample sizes, existing methods are not sufficient. We propose parametric and non-parametric empirical Bayes frameworks for adjusting data for batch effects that is robust to outliers in small sample sizes and performs comparable to existing methods for large samples. We illustrate our methods using two example data sets and show that our methods are justifiable, easy to apply, and useful in practice. Software for our method is freely available at: http://biosun1.harvard.edu/complab/batch/.

Bayes Theorem↗
PubMed2002

Reducing environmental bias when measuring natural selection.

Crucial to understanding the process of natural selection is characterizing phenotypic selection. Measures of phenotypic selection can be biased by environmental variation among individuals that causes a spurious correlation between a trait and fitness. One solution is analyzing genotypic data, rather than phenotypic data. Genotypic data, however, are difficult to gather, can be gathered from few species, and typically have low statistical power. Environmental correlations may act through traits other than through fitness itself. A path analytic framework, which includes measures of such traits, may reduce environmental bias in estimates of selection coefficients. We tested the efficacy of path analysis to reduce bias by re-analyzing three experiments where both phenotypic and genotypic data were available. All three consisted of plant species (Impatiens capensis, Arabidopsis thaliana, and Raphanus sativus) grown in experimental plots or the greenhouse. We found that selection coefficients estimated by path analysis using phenotypic data were highly correlated with those based on genotypic data with little systematic bias in estimating the strength of selection. Although not a panacea, using path analysis can substantially reduce environmental biases in estimates of selection coefficients. Such confidence in phenotypic selection estimates is critical for progress in the study of natural selection.

Arabidopsis↗
PubMed1997

Body size and prostate cancer: a 20-year follow-up study among 135006 Swedish construction workers.

BACKGROUND: Obesity is associated with endocrine changes (e.g., increased estrogen and decreased testosterone in the blood) that have been implicated in the cause of prostate cancer and, therefore, an association between body weight and the risk of developing prostate cancer would be expected. However, because of bias or low statistical power in previous epidemiologic studies, associations between anthropometric measurements (height and weight), body mass index (BMI), and the risk of prostate cancer may have been inadvertently overlooked. PURPOSE: We performed a large, retrospective cohort study among Swedish construction workers to evaluate possible associations of adult weight, height, BMI, and lean body mass (LBM) by age at entry in the study with the incidence and mortality rate of prostate cancer. METHODS: We analyzed data that had been compiled in a computerized central register on a cohort of approximately 135000 male construction workers. Information on height and weight had been collected with the use of a comprehensive questionnaire filled out by nurses at the time of enrollment in the cohort, from 1971 through 1975. Complete follow-up was achieved through 1991 by means of record linkage to the Swedish National Cancer Register, the Death Register, and the Migration Register. A total of 2368 incident cases and 708 deaths from prostate cancer occurred in the cohort during a follow-up period averaging 18 years. We used only information obtained at the index visit from 1971 through 1975 to determine age-adjusted rate ratios (RRs) in a Poisson-based multiplicative multivariate model with age and the relevant exposure variable (e.g., weight, height, BMI, and LBM) as independent variables. RESULTS: All anthropometric measurements were positively associated with the risk of prostate cancer and were more strongly related to mortality than to incidence. The excess risk of death from prostate cancer was statistically significant in all BMI categories above the reference category: RR = 1.40 (95% confidence interval [CI] = 1.09-1.81) in the highest category compared with the lowest (P for trend = .04). For height and LBM, the excess risk in the highest compared with the lowest categories was somewhat less pronounced: RR = 1.28 (95% CI = 1.02-1.60) and RR = 1.26 (95% CI = 1.02-1.57), respectively. Statistically significant linear dose-response relationships were also found with the incidence of prostate cancer, with the exception of BMI (P for trend = .10). CONCLUSION: Our large cohort study indicates that various aspects of body size are related to the risk of prostate cancer and that future studies are needed to study the role of body size and prostate cancer.

Adult↗
PubMed2002

Stapled versus handsewn methods for colorectal anastomosis surgery: a systematic review of randomized controlled trials.

CONTEXT: The interest in the results from comparisons between handsewing and stapling in colorectal anastomoses has been reflected in the progressive increase in the number of clinical trials. These studies, however, do not permit conclusions to be drawn, given the lack of statistical power of the samples analyzed. OBJECTIVE: To compare stapling and handsewing in colorectal anastomosis, testing the hypothesis that in colorectal anastomosis the technique of stapling is superior to that of handsewing. DESIGN: A systematic review of randomized controlled trials. SURVEY STRATEGY: Systematic revision of the literature and meta-analysis were used, without restrictions on language, dates or other considerations. The sources of information used were Embase, Lilacs, Medline, Cochrane Controlled Clinical Trials Database, and letters to authors and industrial producers of staples and thread. SELECTION CRITERIA: Studies were included in accordance with randomization criteria. The external validity of the studies was investigated via the characteristics of the participants, the interventions and the variables analyzed. An independent selection of clinical studies focusing on analysis of adult patients attended to on an elective basis was made by two reviewers. DATA COLLECTION AND ANALYSIS: The methodological quality of the studies was assessed by the same reviewers. In addition to the randomization criteria, the masking, treatment intention, losses and exclusions were also analyzed. The meta-analysis was performed using risk difference and weighted average difference, with their respective 95% confidence intervals. The variables studied were mortality, clinical and radiological anastomotic dehiscence, anastomotic stricture, hemorrhage, reoperation, wound infection, time taken to perform the anastomosis and hospital stay. RESULTS: Nine clinical trials were selected. After verifying that it was possible to perform one of the two techniques being compared, 1,233 patients were included, of whom 622 underwent stapling and 611 the handsewing technique. No statistical difference was found between the variables, except for stenosis, which was more frequent in stapling (p < 0.05), and the time taken to perform the anastomosis, which was greater in handsewing (p < 0.05). CONCLUSION: The evidence found was insufficient to demonstrate superiority of the stapling method over handsewing, independent of the level of colorectal anastomosis.

Anastomosis, Surgical↗
PubMed2001

A family-based strategy to identify genes for diabetic nephropathy.

Diabetic nephropathy (DN) clusters in families and specific ethnic groups, suggesting a genetic basis of disease transmission. Identification of DN susceptibility loci should reveal new therapeutic targets but requires accurate phenotyping. A powerful family-based strategy, which is novel to the pursuit of nephropathy genes in type 2 diabetes, is being used to collect a sample for candidate gene and genome scan analyses. Sib pairs that include DN index cases plus (1) sibs concordant for type 2 diabetes and DN (affected sib pairs [ASPs]) and (2) sibs concordant for type 2 diabetes but discordant for DN (discordant sib pairs [DSPs]) are targeted specifically for recruitment. Type 2 diabetes and DN phenotype criteria for index cases include diabetes onset after 38 years of age, duration 10 years or longer, no initial insulin treatment, diabetic retinopathy, end-stage renal disease (ESRD), and history of nephrotic proteinuria. ESRD patients were screened by questionnaire and medical record review (n = 2114). Of 666 patients with ESRD secondary to DN, 227 had a family history of ESRD, 150 had a living diabetic sib, and 124 families were enrolled. Sixty-five families, with 86 diabetic relative pairs (69 sibs, 17 children), have been completely phenotyped. If nephropathy in diabetic sibs is defined as albuminuria greater than 0.3 g/24 h, 31 ASPs and 26 DSPs (diabetic sib with albuminuria <0.3 g/24 h) were identified. Applying more stringent criteria, only 12 ASPs (sib with diabetes >10 years, diabetic retinopathy, and nephrotic proteinuria) and 9 DSPs (sib with diabetes >10 years and normal urine albumin excretion) were identified. Extrapolating from the number of subjects recruited using stringent phenotyping criteria, nearly 10,000 ESRD patients are required for screening to achieve adequate statistical power for linkage analysis (80% power to detect locus-specific relative risk of 2.2 at a lod score of 3.0). Careful phenotyping requires a large recruitment effort but is necessary to reduce population heterogeneity, a strategy that increases the likelihood of identifying DN loci.

Age of Onset↗
PubMed2000

Performance of population specific job exposure matrices (JEMs): European collaborative analyses on occupational risk factors for chronic obstructive pulmonary disease with job exposure matrices (ECOJEM).

OBJECTIVES: To compare the performance of population specific job exposure matrices (JEMs) and self reported occupational exposure with data on exposure and lung function from three European general populations. METHODS: Self reported occupational exposure (yes or no) and present occupation were recorded in the three general population surveys conducted in France, The Netherlands, and Norway. Analysis was performed on subjects, aged 25-64, who provided good forced expiratory volume in 1 second (FEV1) tracings and whose occupations were performed by at least two people, in the French (6217 men and 5571 women), the Dutch (men from urban (854) and rural (780) areas), and the Norwegian (395 men) surveys. Two population specific JEMs, based on the percentage of subjects who reported themselves exposed in each job, were constructed for each survey and each sex. The first matrix classified jobs into three categories of exposure according to the proportion of subjects who reported themselves exposed in each job (P10-50 JEM, low < 10%, moderate 10-49%, high > or = 50%). For the second matrix, a dichotomous variable was constructed to have the same statistical power as the self reported exposure--that is, the exposure prevalence (p) was the same with both exposure assessment methods (Pp JEM). Relations between occupational exposure, as estimated by the two JEMs and self reported exposure, and age, height, city, and smoking adjusted FEV1 score were compared. RESULTS: Significant associations between occupational exposure estimated by the population specific JEM and lung function were found in the French and the rural Dutch surveys, whereas no significant relation was found with self reported exposure. In populations with few subjects in most jobs, exposure cannot be estimated with sufficient precision by a population specific JEM, which may explain the lack of relation in the Norwegian and the Dutch (urban area) surveys. CONCLUSION: The population specific JEM, which was easy to construct and cost little, seemed to perform better than crude self reported exposures, in populations with sufficient numbers of subjects per job.

Adult↗
PubMed1999

Reduction of myocardial damage by prolonged treatment with subcutaneous low molecular weight heparin in unstable coronary artery disease. FRISC study group. Fragmin during Instability in Coronary Artery Disease.

AIMS: Several studies have proved heparin useful in treating patients with unstable coronary artery disease. The present study investigates whether Selvester QRS scoring for estimation of myocardial infarct size increases the incidence of detection of acute myocardial infarction during follow-up in a trial of patients with unstable angina/non-Q wave myocardial infarction treated with low molecular weight heparin or placebo. Finally it will be discussed how the QRS score, used for end-point identification, impacts on the power calculation in clinical trials. METHODS AND RESULTS: Electrocardiographic data on 1276 patients (644 in the placebo group, 632 in the low molecular weight heparin treatment group) were available. All ECGs were scored according to the Selvester QRS scoring method. At 40 days, more patients in the placebo than in the heparin group had achieved a threshold level of QRS score (25.9% vs 21.1%, P=0.05). Myocardial infarction, diagnosed as per the classic Q wave criteria, occurred in 3.7% of patients in the placebo group and in 0.9% in the low molecular weight heparin group at 6 days (P=0.002). At 40 days, the rates were 8.2% (placebo) and 5.7% (low molecular weight heparin, P=0.2). By combining the classic criteria with the Selvester method the myocardial infarction end-point rate in both groups was almost doubled (8.2% to 14.4% in the placebo group and 5.7% to 11.1% in the low molecular weight heparin group, P=0.07). The 216 patients with non-evaluable electrocardiograms did not differ from the 1276 patients as regards baseline characteristics; however, they had a significantly poorer prognosis, with a death/myocardial infarction rate of 20% at 40 days, compared with 8% among the patients with evaluable electrocardiograms (P<0.00001). CONCLUSION: Long-term subcutaneous treatment with low molecular weight heparin decreases the number of subsequent myocardial infarctions - determined both conventionally and by an increase in QRS score - in patients with unstable coronary artery disease. Silent myocardial infarctions detected by QRS score, as well as clinical myocardial infarctions, could be used as end-points in clinical trials of ischaemic heart disease and thus lower the population needed for obtaining statistical power.

Aged↗
PubMed2003

Optimization of experimental design in fMRI: a general framework using a genetic algorithm.

This article describes a method for selecting design parameters and a particular sequence of events in fMRI so as to maximize statistical power and psychological validity. Our approach uses a genetic algorithm (GA), a class of flexible search algorithms that optimize designs with respect to single or multiple measures of fitness. Two strengths of the GA framework are that (1) it operates with any sort of model, allowing for very specific parameterization of experimental conditions, including nonstandard trial types and experimentally observed scanner autocorrelation, and (2) it is flexible with respect to fitness criteria, allowing optimization over known or novel fitness measures. We describe how genetic algorithms may be applied to experimental design for fMRI, and we use the framework to explore the space of possible fMRI design parameters, with the goal of providing information about optimal design choices for several types of designs. In our simulations, we considered three fitness measures: contrast estimation efficiency, hemodynamic response estimation efficiency, and design counterbalancing. Although there are inherent trade-offs between these three fitness measures, GA optimization can produce designs that outperform random designs on all three criteria simultaneously.

Algorithms↗
PubMed1997

Cerebral and umbilical vascular resistance response to vibroacoustic stimulation in growth-restricted fetuses.

OBJECTIVE: To test the hypothesis that after vibroacoustic stimulation the ratio between cerebral vascular and umbilical vascular resistance in the growth-restricted fetus is different from that in the normal fetus. METHODS: The pulsatility index (PI) of the middle cerebral artery and that of the umbilical artery (UA) were measured by pulsed Doppler velocimetry in 30 normal and 14 growth-restricted fetuses before and after vibroacoustic stimulation. The ratios of cerebral PI to UA PI and the changes in PI after vibroacoustic stimulation were calculated. Comparisons were made using the Wilcoxon rank-sum test or signed-rank test. The statistical power of the study was 80%. RESULTS: Mean (+/- standard deviation) cerebral PI values before vibroacoustic stimulation (1.50 +/- 0.29) in normals and 1.29 +/- 0.26 in the fetal growth restriction [FGR] group) and UA PI values (1.00 +/- 0.18 in normals and 1.15 +/- 0.24 in the FGR group) were significantly different between groups (P < .04) and significantly decreased after vibroacoustic stimulation (P < .05). Although the cerebral to UA PI ratios (1.50 +/- 0.38 in normals and 1.13 +/- 0.33 in the FGR group) were significantly different between groups (P < .008), the values remained the same after vibroacoustic stimulation (P = .39 and .80, respectively). In all fetuses the fetal heart rate accelerated after vibroacoustic stimulation. CONCLUSION: Cerebral vascular resistance was lower and umbilical vascular resistance higher in the growth-restricted fetuses than in normals. The vascular resistance response after vibroacoustic stimulation in the growth-restricted fetus was not significantly different from the response of the normal fetus, suggesting preservation of regulation of resistance.

Acoustic Stimulation↗
PubMed1992

Meta-analysis: the fashion of summing-up evidence. Part II: Interpretations and uses.

In this commentary, we use evidence produced by the Early Breast Cancer Trialists' Collaborative Group (EBCTCG) ten-year update of a meta-analysis of trials of adjuvant therapies for early breast cancer which started prior to 1985 to illustrate aspects of interpretations and uses of meta-analysis results. The following issues are discussed: i) The meta-analysis provides an average summary for the effect of a treatment. Greater statistical power is obtained by increasing the number of events contributing to the analysis. However, summing up the results of various trials necessitates the loss of individual information concerning the magnitude of treatment effects which depend on tumor- and patient-related factors. Subgroup analyses within the meta-analysis process allow some recovery of such features; ii) The absolute benefit obtained from an effective treatment depends not only on the relative benefit of the treatment but also on the prognosis of the individual patients; iii) The results are more immediately applicable if less reliance is placed on the arithmetic construct inherent in the overview, using instead unconfounded information about the value of treatments actually administered. This avoids the need to extrapolate the effect for one component of the therapy by assuming a lack of interaction with its other components; iv) Although indirect comparisons between different meta-analyses are regularly made to pick the "winner" from among tested treatment modalities, it is unlikely that the optimal therapeutic regimen can be defined via such indirect comparisons, though such comparisons may raise interesting, testable hypotheses.

Antineoplastic Combined Chemotherapy Protocols↗
PubMed1999

Population genetics of the yellow fever mosquito in Trinidad: comparisons of amplified fragment length polymorphism (AFLP) and restriction fragment length polymorphism (RFLP) markers.

Recent development of DNA markers provides powerful tools for population genetic analyses. Amplified fragment length polymorphism (AFLP) markers result from a polymerase chain reaction (PCR)-based DNA fingerprinting technique that can detect multiple restriction fragments in a single polyacrylamide gel, and thus are potentially useful for population genetic studies. Because AFLP markers have to be analysed as dominant loci in order to estimate population genetic diversity and genetic structure parameters, one must assume that dominant (amplified) alleles are identical in state, recessive (unamplified) alleles are identical in state, AFLP fragments segregate according to Mendelian expectations and that the genotypes of an AFLP locus are in Hardy-Weinberg equilibrium (HWE). The HWE assumption is untestable for natural populations using dominant markers. Restriction fragment length polymorphism (RFLP) markers segregate as codominant alleles, and can therefore be used to test the HWE assumption that is critical for analysing AFLP data. This study examined whether the dominant AFLP markers could provide accurate estimates of genetic variability for the Aedes aegypti mosquito populations of Trinidad, West Indies, by comparing genetic structure parameters using AFLP and RFLP markers. For AFLP markers, we tested a total of five primer combinations and scored 137 putative loci. For RFLP, we examined a total of eight mapped markers that provide a broad coverage of mosquito genome. The estimated average heterozygosity with AFLP markers was similar among the populations (0.39), and the observed average heterozygosity with RFLP markers varied from 0.44 to 0.58. The average FST (standardized among-population genetic variance) estimates were 0.033 for AFLP and 0.063 for RFLP markers. The genotypes at several RFLP loci were not in HWE, suggesting that the assumption critical for analysing AFLP data was invalid for some loci of the mosquito populations in Trinidad. Therefore, the results suggest that, compared with dominant molecular markers, codominant DNA markers provide better estimates of population genetic variability, and offer more statistical power for detecting population genetic structure.

Aedes↗
PubMed2001

Stapled versus handsewn methods for colorectal anastomosis surgery.

BACKGROUND: Randomized controlled trials comparing stapled with handsewn colorectal anastomosis have not shown either technique to be superior, perhaps because individual studies lacked statistical power. A systematic review, with pooled analysis of results, might provide a more definitive answer. OBJECTIVES: To compare the safety and effectiveness of stapled and handsewn colorectal anastomosis. The following primary hypothesis was tested: the stapled technique is more effective because it decreases the level of complications. SEARCH STRATEGY: The RCT register of the Cochrane Review Group was searched for any trial or reference to a relevant trial (published, in-press, or in progress). All publications were sought through computerised searches of EMBASE, LILACS, MEDLINE, the Cochrane Controlled Clinical Trials Database, and through letters to industrial companies and authors. There were no limits upon language, date, or other criteria. SELECTION CRITERIA STUDIES: All randomized clinical trials (RCTs) in which stapled and handsewn colorectal anastomosis were compared. PARTICIPANTS: Adult patients submitted electively to colorectal anastomosis. INTERVENTIONS: Endoluminal circular stapler and handsewn colorectal anastomosis. OUTCOMES: a) Mortality b) Overall Anastomotic Dehiscence c) Clinical Anastomotic Dehiscence d) Radiological Anastomotic Dehiscence e) Stricture f) Anastomotic Haemorrhage g) Reoperation h) Wound Infection i) Anastomosis Duration j) Hospital Stay. DATA COLLECTION AND ANALYSIS: Data were independently extracted by the two reviewers (SASL, DM) and cross-checked. The methodological quality of each trial was assessed by the same two reviewers. Details of the randomization (generation and concealment), blinding, whether an intention-to-treat analysis was done, and the number of patients lost to follow-up were recorded. The results of each RCT were summarised on an intention-to-treat basis in 2 x 2 tables for each outcome. External validity was defined by characteristics of the participants, the interventions and the outcomes. The RCTs were stratified according to the level of colorectal anastomosis. The Risk Difference method (random effects model) and NNT for dichotomous outcomes measures and weighted mean difference for continuous outcomes measures, with the corresponding 95% confidence interval, were presented in this review. Statistical heterogeneity was evaluated by using funnel plot and chi-square testing. MAIN RESULTS: Of the 1233 patients enrolled ( in 9 trials), 622 were treated with stapled, and 611 with manual, suture. The following main results were obtained: a) Mortality: result based on 901 patients; Risk Difference - 0.6% Confidence Interval -2.8% to +1.6%. b) Overall Dehiscence: result based on 1233 patients; Risk Difference 0.2%, 95% Confidence Interval -5.0% to +5.3%. c) Clinical Anastomotic Dehiscence : result based on 1233 patients; Risk Difference -1.4%, 95% Confidence Interval -5.2 to +2.3%. d) Radiological Anastomotic Dehiscence : result based on 825 patients; Risk Difference 1.2%, 95% Confidence Interval -4.8% to +7.3%. e) Stricture: result based on 1042 patients; Risk Difference 4.6%, 95% Confidence Interval 1.2% to 8.1%. Number needed to treat 17, 95% confidence interval 12 to 31. f) Anastomotic Hemorrhage: result based on 662 patients; Risk Difference 2.7%, 95% Confidence Interval - 0.1% to +5.5%. g) Reoperation: result based on 544 patients; Risk Difference 3.9%, 95% Confidence Interval 0.3% to 7.4%. h) Wound Infection: result based on 567 patients; Risk Difference 1.0%, 95% Confidence Interval -2.2% to +4.3%. i) Anastomosis duration: result based on one study (159 patients); Weighted Mean Difference -7.6 minutes, 95% Confidence Interval -12.9 to -2.2 minutes. j) Hospital Stay: result based on one study (159 patients), Weighted Mean Difference 2.0 days, 95% Confidence Interval -3.27 to +7.2 days. REVIEWER'S CONCLUSIONS: The evidence found was insufficient to demonstrate any superiority of stapled over handsewn techniques in colorectal anastomosis, regardless of the level of anastomosis.

Adult↗
PubMedDate not supplied

A criterion measurement model for health behavior change.

Researchers in the field of health behavior change have traditionally relied on a univariate criterion measure to evaluate the efficacy of an intervention. Such measures have superficial face validity but suffer from a number of problems: (a) lack of precise definitions; (b) poor statistical power; and (c) a lack of meaningfulness for some aspects of the problem. As an alternative, a theoretical model is developed that attempts to define more appropriate multivariate sets of dependent variables for the study of health behavior change. The model involves three separate constructs: Positive Evaluation Strength, Negative Evaluation Strength, and Habit Strength. The pattern of change for each construct is described across four stages of change: Precontemplation, Contemplation, Action, and Maintenance. For each construct, two thresholds are proposed representing the ability of the environment to modify the construct. Four tests of the model are provided from existing data sets. First, a structural model analysis was used to test if the proposed measurement model adequately fits the data. Second, a dynamic typology approach produced profiles of change that are consistent with the model. Third, a time series analysis provided support for the assumed model. Fourth, longitudinal, five-wave panel design was employed to test if the relation between the two cognitive variables (Pros and Cons) and the behavioral measure (Habit Strength) was consistent with the model. Implications for alternative intervention strategies are discussed.

Behavior, Addictive↗
PubMed1997

Re-using data from case-control studies.

Despite its ability to maximize statistical power while keeping data collection costs to a minimum, case-control sampling provides a non-representative sample of the population. When fitting a logistic regression model to data obtained in such a study, using the variable stratifying the population as the response, it is well known that the estimate of the constant term will be biased, but those of the coefficients of the covariates will not. However, subsequent to the case-control study, it is often desired to conduct a secondary analysis, using a variable that was previously a covariate in the main study as the response. If this new response is associated with the original variable used to stratify the population into cases and controls, a conventional logistic regression analysis will usually result in biased estimates of all the regression coefficients, not just the constant. This situation has recently been studied by Nagelkerke et al. who describe some situations where no bias occurs. In this paper we discuss how to calculate maximum likelihood estimates of all the regression coefficients, in the situation where the sampling rates for cases and controls are known. An example using data from the New Zealand Cot Death Study is presented.

Case-Control Studies↗
PubMed2000

Multivariate genetic analysis of brain structure in an extended twin design.

The hunt for genes influencing behavior may be aided by the study of intermediate phenotypes for several reasons. First, intermediate phenotypes may be influenced by only a few genes, which facilitates their detection. Second, many intermediate phenotypes can be measured on a continuous quantitative scale and thus can be assessed in affected and unaffected individuals. Continuous measures increase the statistical power to detect genetic effects (Neale et al., 1994), and allow studies to be designed to collect data from informative subjects such as extreme concordant or discordant pairs. Intermediate phenotypes for discrete traits, such as psychiatric disorders, can be neurotransmitter levels, brain function, or structure. In this paper we conduct a multivariate analysis of data from 111 twin pairs and 34 additional siblings on cerebellar volume, intracranial space, and body height. The analysis is carried out on the raw data and specifies a model for the mean and the covariance structure. Results suggest that cerebellar volume and intracranial space vary with age and sex. Brain volumes tend to decrease slightly with age, and males generally have a larger brain volume than females. The remaining phenotypic variance of cerebellar volume is largely genetic (88%). These genetic factors partly overlap with the genetic factors that explain variance in intracranial space and body height. The applied method is presented as a general approach for the analysis of intermediate phenotypes in which the effects of correlated variables on the observed scores are modeled through multivariate analysis.

Adult↗
PubMed2004

Clinical trials and p-values, beware of the extremes.

BACKGROUND: In randomized controlled trials, prior to statistical analysis, the data are checked for outliers and erroneous data. Statistical tests are, traditionally, not very good at distinguishing between errors and outliers, but they should be able to point out main endpoint results closer to expectation than compatible with random sampling. OBJECTIVES: To explain from hypothesized and published examples why extreme p-values like p>0.95 and p<0.0001 may indicate that sampling was not completely random. RESULTS: Extreme p-values can be readily observed in recent issues of high-impact journals. A p-value >0.95 literally means that we have a >95% chance of finding a result less close to expectation and, consequently, a <5% chance of finding a result this close or closer. Often in studies a statistical power of 80% is agreed upon, corresponding with a p-value of approximately 0.01. The ultimate p-value may then be a bit larger or smaller. However, a p-value much smaller than 0.01 will be rarely observed, because it would indicate that the study is overpowered. If the p-values can be assumed to follow a normal distribution around 0.01, then we will have a less than 5% chance of observing a p-value of <0.0001. CONCLUSIONS: In randomized controlled trials, main endpoint p-values larger than p=0.95 will be rare, because they would indicate similarities closer than compatible with a normal distribution of random data samples. Also very low p-values like p<0.0001 will be rarely encountered, because it would mean that the trial was overpowered and should have had a smaller sample size. It would seem appropriate, therefore, to require investigators to explain such results and to consider rejecting the research involved. So far, in randomized controlled trials the null-hypothesis is generally rejected at p<0.05. Perhaps we should consider rejecting the entire study if the main endpoint p-values are >0.95 or <0.0001.

Area Under Curve↗