PMCC PMCC

Search tips
Search criteria

Advanced
Results 1-8 (8)
 

Clipboard (0)
None

Select a Filter Below

Journals
Year of Publication
author:("lake, Petter")
1.  The McNemar test for binary matched-pairs data: mid-p and asymptotic are better than exact conditional 
Background
Statistical methods that use the mid-p approach are useful tools to analyze categorical data, particularly for small and moderate sample sizes. Mid-p tests strike a balance between overly conservative exact methods and asymptotic methods that frequently violate the nominal level. Here, we examine a mid-p version of the McNemar exact conditional test for the analysis of paired binomial proportions.
Methods
We compare the type I error rates and power of the mid-p test with those of the asymptotic McNemar test (with and without continuity correction), the McNemar exact conditional test, and an exact unconditional test using complete enumeration. We show how the mid-p test can be calculated using eight standard software packages, including Excel.
Results
The mid-p test performs well compared with the asymptotic, asymptotic with continuity correction, and exact conditional tests, and almost as good as the vastly more complex exact unconditional test. Even though the mid-p test does not guarantee preservation of the significance level, it did not violate the nominal level in any of the 9595 scenarios considered in this article. It was almost as powerful as the asymptotic test. The exact conditional test and the asymptotic test with continuity correction did not perform well for any of the considered scenarios.
Conclusions
The easy-to-calculate mid-p test is an excellent alternative to the complex exact unconditional test. Both can be recommended for use in any situation. We also recommend the asymptotic test if small but frequent violations of the nominal level is acceptable.
doi:10.1186/1471-2288-13-91
PMCID: PMC3716987  PMID: 23848987
Matched pairs; Dependent proportions; Paired proportions; Quasi-exact
2.  Comparing hospital mortality – how to count does matter for patients hospitalized for acute myocardial infarction (AMI), stroke and hip fracture 
Background
Mortality is a widely used, but often criticised, quality indicator for hospitals. In many countries, mortality is calculated from in-hospital deaths, due to limited access to follow-up data on patients transferred between hospitals and on discharged patients. The objectives were to: i) summarize time, place and cause of death for first time acute myocardial infarction (AMI), stroke and hip fracture, ii) compare case-mix adjusted 30-day mortality measures based on in-hospital deaths and in-and-out-of hospital deaths, with and without patients transferred to other hospitals.
Methods
Norwegian hospital data within a 5-year period were merged with information from official registers. Mortality based on in-and-out-of-hospital deaths, weighted according to length of stay at each hospital for transferred patients (W30D), was compared to a) mortality based on in-and-out-of-hospital deaths excluding patients treated at two or more hospitals (S30D), and b) mortality based on in-hospital deaths (IH30D). Adjusted mortalities were estimated by logistic regression which, in addition to hospital, included age, sex and stage of disease. The hospitals were assigned outlier status according to the Z-values for hospitals in the models; low mortality: Z-values below the 5-percentile, high mortality: Z-values above the 95-percentile, medium mortality: remaining hospitals.
Results
The data included 48 048 AMI patients, 47 854 stroke patients and 40 142 hip fracture patients from 55, 59 and 58 hospitals, respectively. The overall relative frequencies of deaths within 30 days were 19.1% (AMI), 17.6% (stroke) and 7.8% (hip fracture). The cause of death diagnoses included the referral diagnosis for 73.8-89.6% of the deaths within 30 days. When comparing S30D versus W30D outlier status changed for 14.6% (AMI), 15.3% (stroke) and 36.2% (hip fracture) of the hospitals. For IH30D compared to W30D outlier status changed for 18.2% (AMI), 25.4% (stroke) and 27.6% (hip fracture) of the hospitals.
Conclusions
Mortality measures based on in-hospital deaths alone, or measures excluding admissions for transferred patients, can be misleading as indicators of hospital performance. We propose to attribute the outcome to all hospitals by fraction of time spent in each hospital for patients transferred between hospitals to reduce bias due to double counting or exclusion of hospital stays.
doi:10.1186/1472-6963-12-364
PMCID: PMC3526398  PMID: 23088745
Mortality; Quality indicator; Transferred patients; AMI; Stroke; Hip fracture; Cause of death; Hospital comparison; Episode of care
4.  Categorisation of continuous exposure variables revisited. A response to the Hyperglycaemia and Adverse Pregnancy Outcome (HAPO) Study 
Background
Although the general statistical advice is to keep continuous exposure variables as continuous in statistical analyses, categorisation is still a common approach in medical research. In a recent paper from the Hyperglycaemia and Adverse Pregnancy Outcome (HAPO) Study, categorisation of body mass index (BMI) was used when analysing the effect of BMI on adverse pregnancy outcomes. The lowest category, labelled "underweight", was used as the reference category.
Methods
The present paper gives a summary of reasons for categorisation and methodological drawbacks of this approach. We also discuss the choice of reference category and alternative analyses. We exemplify our arguments by a reanalysis of results from the HAPO paper.
Results
Categorisation of continuous exposure data results in loss of power and other methodological challenges. An unfortunate choice of reference category can give additional lack of precision and obscure the interpretation of risk estimates. A highlighted odds ratio (OR) in the HAPO study is the OR for birth weight >90th percentile for women in the highest compared to the lowest BMI category ("obese class III" versus "underweight"). This estimate was OR = 4.55 and OR = 3.52, with two different multiple logistic regression models. When using the "normal weight" category as the reference, our corresponding estimates were OR = 2.03 and OR = 1.62, respectively. Moreover, our choice of reference category also gave narrower confidence intervals.
Summary
Due to several methodological drawbacks, categorisation should be avoided. Modern statistical analyses should be used to analyse continuous exposure data, and to explore non-linear relations. If continuous data are categorised, special attention must be given to the choice of reference category.
doi:10.1186/1471-2288-10-103
PMCID: PMC2992539  PMID: 21062456
5.  Blood cell gene expression associated with cellular stress defense is modulated by antioxidant-rich food in a randomised controlled clinical trial of male smokers 
BMC Medicine  2010;8:54.
Background
Plant-based diets rich in fruit and vegetables can prevent development of several chronic age-related diseases. However, the mechanisms behind this protective effect are not elucidated. We have tested the hypothesis that intake of antioxidant-rich foods can affect groups of genes associated with cellular stress defence in human blood cells. Trial registration number: NCT00520819 http://clinicaltrials.gov.
Methods
In an 8-week dietary intervention study, 102 healthy male smokers were randomised to either a diet rich in various antioxidant-rich foods, a kiwifruit diet (three kiwifruits/d added to the regular diet) or a control group. Blood cell gene expression profiles were obtained from 10 randomly selected individuals of each group. Diet-induced changes on gene expression were compared to controls using a novel application of the gene set enrichment analysis (GSEA) on transcription profiles obtained using Affymetrix HG-U133-Plus 2.0 whole genome arrays.
Results
Changes were observed in the blood cell gene expression profiles in both intervention groups when compared to the control group. Groups of genes involved in regulation of cellular stress defence, such as DNA repair, apoptosis and hypoxia, were significantly upregulated (GSEA, FDR q-values < 5%) by both diets compared to the control group. Genes with common regulatory motifs for aryl hydrocarbon receptor (AhR) and AhR nuclear translocator (AhR/ARNT) were upregulated by both interventions (FDR q-values < 5%). Plasma antioxidant biomarkers (polyphenols/carotenoids) increased in both groups.
Conclusions
The observed changes in the blood cell gene expression profiles suggest that the beneficial effects of a plant-based diet on human health may be mediated through optimization of defence processes.
doi:10.1186/1741-7015-8-54
PMCID: PMC2955589  PMID: 20846424
7.  Regression analysis with categorized regression calibrated exposure: some interesting findings 
Background
Regression calibration as a method for handling measurement error is becoming increasingly well-known and used in epidemiologic research. However, the standard version of the method is not appropriate for exposure analyzed on a categorical (e.g. quintile) scale, an approach commonly used in epidemiologic studies. A tempting solution could then be to use the predicted continuous exposure obtained through the regression calibration method and treat it as an approximation to the true exposure, that is, include the categorized calibrated exposure in the main regression analysis.
Methods
We use semi-analytical calculations and simulations to evaluate the performance of the proposed approach compared to the naive approach of not correcting for measurement error, in situations where analyses are performed on quintile scale and when incorporating the original scale into the categorical variables, respectively. We also present analyses of real data, containing measures of folate intake and depression, from the Norwegian Women and Cancer study (NOWAC).
Results
In cases where extra information is available through replicated measurements and not validation data, regression calibration does not maintain important qualities of the true exposure distribution, thus estimates of variance and percentiles can be severely biased. We show that the outlined approach maintains much, in some cases all, of the misclassification found in the observed exposure. For that reason, regression analysis with the corrected variable included on a categorical scale is still biased. In some cases the corrected estimates are analytically equal to those obtained by the naive approach. Regression calibration is however vastly superior to the naive method when applying the medians of each category in the analysis.
Conclusion
Regression calibration in its most well-known form is not appropriate for measurement error correction when the exposure is analyzed on a percentile scale. Relating back to the original scale of the exposure solves the problem. The conclusion regards all regression models.
doi:10.1186/1742-7622-3-6
PMCID: PMC1559617  PMID: 16820052
8.  Test-retest reproducibility of a food frequency questionnaire (FFQ) and estimated effects on disease risk in the Norwegian Women and Cancer Study (NOWAC) 
Nutrition Journal  2006;5:4.
Background
The Norwegian Women and Cancer Study (NOWAC) is a national population-based cohort study with 102 443 women enrolled at age 30–70 y from 1991 to 1997. The present study was a methodological sub-study to assess the test-retest reproducibility of the NOWAC food frequency questionnaire (FFQ), and to study how measurement errors in the data can affect estimates of disease risk.
Methods
A random sample of 2000 women aged 46–75 y was drawn from the cohort in 2002. A self-instructive health and lifestyle questionnaire with a FFQ section was mailed to the same subjects twice (test-retest), about three months apart, with a response rate of 75%. The FFQ was designed to assess habitual diet over the past year. We assess the reproducibility of single questions, food groups, energy, and nutrients with several statistical measures. We also demonstrate the method of regression calibration to correct disease risk estimates for measurement error. Alcohol intake (g/day) and high blood pressure (yes/no) is used in the example.
Results
For single foods there were some indications of seasonal reporting bias. For food groups and nutrients the reliability coefficients ranged from 0.5–0.8, and Pearson's r, Spearman's rs, and two intraclass correlation coefficients gave similar results. Although alcohol intake had relatively high reproducibility (r = 0.72), odds ratio estimates for the association with blood pressure were attenuated towards the null value compared to estimates corrected by regression calibration.
Conclusion
The level of reproducibility observed for the FFQ used in the NOWAC study is within the range reported for similar instruments, but may attenuate estimates of disease risk.
doi:10.1186/1475-2891-5-4
PMCID: PMC1434764  PMID: 16448553

Results 1-8 (8)