Reflection on modern methods: Statistics education beyond 'significance': novel plain English interpretations to deepen understanding of statistics and to steer away from misinterpretations.

Concerns have been expressed over standards of statistical interpretation. Results with P <0.05 are often referred to as 'significant' which, in plain English, implies important. This leads some people directly into the misconception that this provides proof that associations are clinically relevant. There are calls for statistics educators to respond to these concerns. This article provides novel plain English interpretations that are designed to deepen understanding. Experience teaching postgraduates at Imperial College is discussed. A key issue with focusing on 'significance' is the common inappropriate practice of implying no association exists, simply because P >0.05. Referring to strengths of association in 'study participants' gives them gravitas, which may help to avoid this. This contrasts with the common practice of focusing on imprecision, by referring to the 'sample' and to 'point estimates'. Unlike formal statistical definitions, interpretations developed and presented here are rooted in the application of statistics. They are based on one set of study participants (not many random samples). Precision of strengths of association are based on using strengths in study participants to estimate strengths of association in the population (from which participants were selected by probability random sampling). Reference to 'compatibility with study data, dependent on statistical modelling assumptions' reminds us of the importance of data quality and modelling assumptions. A straightforward graph shows the relationship between P-values and test statistics. This figure and associated interpretations were developed to illuminate the continuous nature of P-values. This is designed to discourage focus on whether P <0.05, and encourage interpretation of exact P-values.

Contract and ownership type of general practices and patient experience in England: multilevel analysis of a national cross-sectional survey.

Objective To examine associations between the contract and ownership type of general practices and patient experience in England. Design Multilevel linear regression analysis of a national cross-sectional patient survey (General Practice Patient Survey). Setting All general practices in England in 2013-2014 ( n = 8017). Participants 903,357 survey respondents aged 18 years or over and registered with a general practice for six months or more (34.3% of 2,631,209 questionnaires sent). Main outcome measures Patient reports of experience across five measures: frequency of consulting a preferred doctor; ability to get a convenient appointment; rating of doctor communication skills; ease of contacting the practice by telephone; and overall experience (measured on four- or five-level interval scales from 0 to 100). Models adjusted for demographic and socioeconomic characteristics of respondents and general practice populations and a random intercept for each general practice. Results Most practices had a centrally negotiated contract with the UK government ('General Medical Services' 54.6%; 4337/7949). Few practices were limited companies with locally negotiated 'Alternative Provider Medical Services' contracts (1.2%; 98/7949); these practices provided worse overall experiences than General Medical Services practices (adjusted mean difference -3.04, 95% CI -4.15 to -1.94). Associations were consistent in direction across outcomes and largest in magnitude for frequency of consulting a preferred doctor (-12.78, 95% CI -15.17 to -10.39). Results were similar for practices owned by large organisations (defined as having ≥20 practices) which were uncommon (2.2%; 176/7949). Conclusions Patients registered to general practices owned by limited companies, including large organisations, reported worse experiences of their care than other patients in 2013-2014.

The Association Between Chronic Pain and Cardiac Disease: A Cross-sectional Population Study.

OBJECTIVES: Chronic pain may increase the risk of cardiac disease, but the extent to which confounding variables account for this association has yet to be satisfactorily established. This study aims to examine the possibility of an independent association between these 2 variables. METHODS: We applied logistic regression analysis to data from 8596 adults surveyed in a population study of the health of the population of England. The association between cardiac disease (angina and/or myocardial infarction) and chronic pain (pain lasting >3 months) was explored, taking account of 10 potentially confounding variables including the regular use of nonsteroidal anti-inflammatory drugs. RESULTS: Participants reporting chronic pain (n=3023) were more likely to experience cardiac disease than those without pain: odds ratio (OR), 1.55; 95% confidence interval (CI), 1.15-2.07. Subsets of participants fulfilling various criteria for high-intensity chronic pain demonstrated stronger associations with cardiac disease suggesting a "dose-response" element to the relationship: chronic widespread pain (OR, 3.3; 95% CI, 1.42-7.68); higher-disability chronic pain (OR, 2.35; 95% CI, 1.71-3.23); and higher average chronic pain score (OR, 1.95; 95% CI, 1.40-2.71). Adjustment for regular prescription of nonsteroidal anti-inflammatory drugs did not reduce the association of chronic pain with cardiac disease. DISCUSSION: Patients reporting chronic pain, in particular those most severely affected, may be at significantly increased risk of cardiac disease. Future studies should focus on determining whether reducing the impact of chronic pain can improve cardiac health.

A Metabolic Study of Huntington's Disease.

BACKGROUND: Huntington's disease patients have a number of peripheral manifestations suggestive of metabolic and endocrine abnormalities. We, therefore, investigated a number of metabolic factors in a 24-hour study of Huntington's disease gene carriers (premanifest and moderate stage II/III) and controls. METHODS: Control (n = 15), premanifest (n = 14) and stage II/III (n = 13) participants were studied with blood sampling over a 24-hour period. A battery of clinical tests including neurological rating and function scales were performed. Visceral and subcutaneous adipose distribution was measured using magnetic resonance imaging. We quantified fasting baseline concentrations of glucose, insulin, cholesterol, triglycerides, lipoprotein (a), fatty acids, amino acids, lactate and osteokines. Leptin and ghrelin were quantified in fasting samples and after a standardised meal. We assessed glucose, insulin, growth hormone and cortisol concentrations during a prolonged oral glucose tolerance test. RESULTS: We found no highly significant differences in carbohydrate, protein or lipid metabolism markers between healthy controls, premanifest and stage II/III Huntington's disease subjects. For some markers (osteoprotegerin, tyrosine, lysine, phenylalanine and arginine) there is a suggestion (p values between 0.02 and 0.05) that levels are higher in patients with premanifest HD, but not moderate HD. However, given the large number of statistical tests performed interpretation of these findings must be cautious. CONCLUSIONS: Contrary to previous studies that showed altered levels of metabolic markers in patients with Huntington's disease, our study did not demonstrate convincing evidence of abnormalities in any of the markers examined. Our analyses were restricted to Huntington's disease patients not taking neuroleptics, anti-depressants or other medication affecting metabolic pathways. Even with the modest sample sizes studied, the lack of highly significant results, despite many being tested, suggests that the majority of these markers do not differ markedly by disease status.

A 24-Hour Study of the Hypothalamo-Pituitary Axes in Huntington's Disease.

BACKGROUND: Huntington's disease is an inherited neurodegenerative disorder characterised by motor, cognitive and psychiatric disturbances. Patients exhibit other symptoms including sleep and mood disturbances, muscle atrophy and weight loss which may be linked to hypothalamic pathology and dysfunction of hypothalamo-pituitary axes. METHODS: We studied neuroendocrine profiles of corticotropic, somatotropic and gonadotropic hypothalamo-pituitary axes hormones over a 24-hour period in controlled environment in 15 healthy controls, 14 premanifest and 13 stage II/III Huntington's disease subjects. We also quantified fasting levels of vasopressin, oestradiol, testosterone, dehydroepiandrosterone sulphate, thyroid stimulating hormone, free triiodothyronine, free total thyroxine, prolactin, adrenaline and noradrenaline. Somatotropic axis hormones, growth hormone releasing hormone, insulin-like growth factor-1 and insulin-like factor binding protein-3 were quantified at 06:00 (fasting), 15:00 and 23:00. A battery of clinical tests, including neurological rating and function scales were performed. RESULTS: 24-hour concentrations of adrenocorticotropic hormone, cortisol, luteinizing hormone and follicle-stimulating hormone did not differ significantly between the Huntington's disease group and controls. Daytime growth hormone secretion was similar in control and Huntington's disease subjects. Stage II/III Huntington's disease subjects had lower concentration of post-sleep growth hormone pulse and higher insulin-like growth factor-1:growth hormone ratio which did not reach significance. In Huntington's disease subjects, baseline levels of hypothalamo-pituitary axis hormones measured did not significantly differ from those of healthy controls. CONCLUSIONS: The relatively small subject group means that the study may not detect subtle perturbations in hormone concentrations. A targeted study of the somatotropic axis in larger cohorts may be warranted. However, the lack of significant results despite many variables being tested does imply that the majority of them do not differ substantially between HD and controls.

Use and abuse of statistics in tobacco industry-funded research on standardised packaging.

In this commentary we consider the validity of tobacco industry-funded research on the effects of standardised packaging in Australia. As the first country to introduce standardised packs, Australia is closely watched, and Philip Morris International has recently funded two studies into the impact of the measure on smoking prevalence. Both of these papers are flawed in conception as well as design but have nonetheless been widely publicised as cautionary tales against standardised pack legislation. Specifically, we focus on the low statistical significance of the analytical methods used and the assumption that standardised packaging should have an immediate large impact on smoking prevalence.

Impact of a COPD discharge care bundle on readmissions following admission with acute exacerbation: interrupted time series analysis.

OBJECTIVES: We evaluated the impact of a COPD discharge care bundle on readmission rates following hospitalisation with an acute exacerbation. DESIGN: Interrupted time series analysis, comparing readmission rates for COPD exacerbations at nine trusts that introduced the bundle, to two comparison groups; (1) other NHS trusts in London and (2) all other NHS trusts in England. Care bundles were implemented at different times for different NHS trusts, ranging from October 2009 to April 2011. SETTING: Nine NHS acute trusts in the London, England. PARTICIPANTS: Patients aged 45 years and older admitted to an NHS acute hospital in England for acute exacerbation of COPD. Data come from Hospital Episode Statistics, April 2002 to March 2012. MAIN OUTCOME MEASURES: Annual trend readmission rates (and in total bed days) within 7, 28 and 90 days, before and after implementation. RESULTS: In hospitals introducing the bundle readmission rates were rising before implementation and falling afterwards (e.g. readmissions within 28 days +2.13% per annum (pa) pre and -5.32% pa post (p for difference in trends = 0.012)). Following implementation, readmission rates within 7 and 28 day were falling faster than among other trusts in London, although this was not statistically significant (e.g. readmissions within 28 days -4.6% pa vs. -3.2% pa, p = 0.44). Comparisons with a national control group were similar. CONCLUSIONS: The COPD discharge care bundle appeared to be associated with a reduction in readmission rate among hospitals using it. The significance of this is unclear because of changes to background trends in London and nationally.

Access to general practice and visits to accident and emergency departments in England: cross-sectional analysis of a national patient survey.

BACKGROUND: The annual number of unplanned attendances at accident and emergency (A&E) departments in England increased by 11% (2.2 million attendances) between 2008-2009 and 2012-2013. A national review of urgent and emergency care has emphasised the role of access to primary care services in preventing A&E attendances. AIM: To estimate the number of A&E attendances in England in 2012-2013 that were preceded by the attending patient being unable to obtain an appointment or a convenient appointment at their general practice. DESIGN AND SETTING: Cross-sectional analysis of a national survey of adults registered with a GP in England. METHOD: The number of general practice consultations in England in 2012-2013 was estimated by extrapolating the linear trend of published data for 2000-2001 to 2008-2009. This parameter was multiplied by the ratio of attempts to obtain a general practice appointment that resulted in an A&E attendance to attempts that resulted in a general practice consultation estimated using the GP Patient Survey 2012-2013. A sensitivity analysis varied the number of consultations by ±12% and the ratio by ±25%. RESULTS: An estimated 5.77 million (99.9% confidence interval = 5.49 to 6.05 million) A&E attendances were preceded by the attending patient being unable to obtain a general practice appointment or a convenient appointment, comprising 26.5% of unplanned A&E attendances in England in 2012-2013. The sensitivity analysis produced values between 17.5% and 37.2% of unplanned A&E attendances. CONCLUSION: A large number of A&E attendances are likely to be preceded by unsuccessful attempts to obtain convenient general practice appointments in England each year.

Influence of area and individual lifecourse deprivation on health behaviours: findings from the British Women's Heart and Health Study.

BACKGROUND: Variable findings have been reported on the contribution of census-based measures of area deprivation over and above that of individual socioeconomic position (SEP) on health outcomes. This study aims to examine the association between residence in a deprived area and health behaviours (diet, smoking and physical inactivity), and how this association is influenced by lifecourse SEP of individuals. DESIGN: A population-based longitudinal study of women aged 60-79 years in 1999-2001 recruited from one general practice in each of 23 British towns. METHODS: Three thousand five hundred twenty-two women were included in the analyses. Area deprivation scores were derived from postcode for residence and lifecourse SEP scores were calculated using 10 individual level indicators of childhood and adult circumstances. To allow direct comparisons of effect of area deprivation and individual SEP, we standardized both measures by generating relative indices of inequality. RESULTS: Both area deprivation and lifecourse SEP were independent predictors of eating fruit and vegetables [odds ratio (OR): 2.87, 95% confidence interval (CI): 2.22-3.72; comparing highest with lowest area Index of Multiple Deprivation of inequality (OR: 3.07, 95% CI: 2.33-4.06) for lifecourse SEP index of inequality] and exercise habits (OR: 2.39, 95% CI: 1.86-3.06 area deprivation; OR: 2.7, 95% CI: 2.07-3.51 individual SEP). Area deprivation was a stronger predictor of smoking behaviour (OR: 2.34, 95% CI: 1.91-3.08) than individual lifecourse SEP (OR: 1.51, 95% CI: 1.17-1.95). CONCLUSION: Most health behaviours among older women were independently associated with both living in deprived areas and individual lifecourse SEP. This suggests that additional health promotion approaches focusing on improving environments would have potential to improve health behaviour.

Influence of life course socioeconomic position on older women's health behaviors: findings from the British Women's Heart and Health Study.

OBJECTIVES: We examined the association between health behaviors and socioeconomic status (SES) in childhood and adult life. METHODS: Self-reported diet, smoking, and physical activity were determined among 3523 women aged 60 to 79 years recruited from general practices in 23 British towns from 1999 through 2001. RESULTS: The most affluent women reported eating more fruit, vegetables, chicken, and fish and less red or processed meat than did less affluent women. Affluent women were less likely to smoke and more likely to exercise. Life course SES did not influence the types of fat, bread, and milk consumed. Adult SES predicted consumption of all foods considered and predicted smoking and physical activity habits independently of childhood SES. Childhood SES predicted fruit and vegetable consumption independently of adult SES and, to a lesser extent, predicted physical activity. Downward social mobility over the life course was associated with poorer diets and reduced physical activity. CONCLUSIONS: Among older women, healthful eating and physical activity were associated with both current and childhood SES. Interventions designed to improve social inequalities in health behaviors should be applied during both childhood and adult life.

Duration of untreated psychosis and social function: 1-year follow-up study of first-episode schizophrenia.

BACKGROUND: In first-episode schizophrenia, longer duration of untreated psychosis (DUP) predicts poorer outcomes. AIMS: To address whether the relationship between DUP and outcome is a direct causal one or the result of association between symptoms and/or cognitive functioning and social functioning at the same time point. METHOD: Symptoms, social function and cognitive function were assessed in 98 patients with first-episode schizphrenia at presentation and 1 year later. RESULTS: There was no significant clinical difference between participants with short and long DUP at presentation. Linear regression analyses revealed that longer DUP significantly predicted more severe positive and negative symptoms and poorer social function at 1 year, independent of scores at presentation. Path analyses revealed independent direct relationships between DUP and social function, core negative symptoms and positive symptoms. There was no significant association between DUP and cognition. CONCLUSIONS: Longer DUP predicts poor social function independently of symptoms. The findings underline the importance of taking account of the phenomenological overlap between measures of negative symptoms and social function when investigating the effects of DUP.

Comorbid substance use and age at onset of schizophrenia.

BACKGROUND: Substance use may be a risk factor for the onset of schizophrenia. AIMS: To examine the association between substance use and age at onset in substance use and age at onset in a UK, inner-city sample of people with recent-onset schizophrenia. METHOD: The study sample consisted of 152 people recruited to the West London First-Episode Schizophrenia Study. Self-reported data on drug and alcohol use, as well as information on age at onset of psychosis, were collected. Mental state, cognition (IQ, memory and executive function) and social function were also assessed. RESULTS: In total, 60% of the participants were smokers, 27% reported a history of problems with alcohol use, 35% reported current substance use (not including alcohol), and 68% reported lifetime substance use (cannabis and psychostimulants were most commonly used). Cannabis use and gender had independent effects on age at onset of psychosis, after adjusting for alcohol misuse and use of other drugs. CONCLUSIONS: The strong association between self-reported cannabis use and earlier onset of psychosis provides further evidence that schizophrenia may be precipitated by cannabis use and/or that the early onset of symptoms is a risk factor for cannabis use.

A volumetric magnetic resonance imaging study of the amygdala in frontotemporal lobar degeneration and Alzheimer's disease.

The amygdala is severely atrophied at post-mortem in frontotemporal lobar degeneration (FTLD), and may contribute to the prominent behavioural changes that are early features of FTLD. The aim of this study was to assess amygdala atrophy using MRI in the main syndromic variants of FTLD and Alzheimer's disease (AD). Brain and amygdala volumes, adjusted for intracranial volume, were measured on 46 clinically diagnosed FTLD patients [22 frontal variant FTD (FTD), 14 semantic dementia (SD), 10 progressive non-fluent aphasia (PNFA)], 20 AD patients, and 17 controls. While severe amygdala atrophy was present in both FTLD (41% smaller than controls on the left; 33% on the right) and in AD (22% on the left; 19% on the right), the FTLD group had significantly greater amygdala atrophy (z = 3.21, p = 0.001 left, z = 2.50, p = 0.01 right) and left/right asymmetry (z = 2.03, p = 0.04) than AD. Amygdala atrophy was greater in SD than FTD, PNFA and AD (p < 0.02 for all). Highly asymmetrical atrophy was present in SD, greater on the left (z = 3.23, p = 0.001), and to a lesser extent in PNFA. Despite an overlap between clinical and radiological features of FTLD and AD, marked amygdala atrophy points towards a diagnosis of FTLD, with left greater than right atrophy suggestive of one of the language variants.

Characteristics of two distinct clinical phenotypes in pathologically proven progressive supranuclear palsy: Richardson's syndrome and PSP-parkinsonism.

The clinical diagnosis of progressive supranuclear palsy (PSP) relies on the identification of characteristic signs and symptoms. A proportion of pathologically diagnosed cases do not develop these classic features, prove difficult to diagnose during life and are considered as atypical PSP. The aim of this study was to examine the apparent clinical dichotomy between typical and atypical PSP, and to compare the biochemical and genetic characteristics of these groups. In 103 consecutive cases of pathologically confirmed PSP, we have identified two clinical phenotypes by factor analysis which we have named Richardson's syndrome (RS) and PSP-parkinsonism (PSP-P). Cases of RS syndrome made up 54% of all cases, and were characterized by the early onset of postural instability and falls, supranuclear vertical gaze palsy and cognitive dysfunction. A second group of 33 (32%) were characterized by asymmetric onset, tremor, a moderate initial therapeutic response to levodopa and were frequently confused with Parkinson's disease (PSP-P). Fourteen cases (14%) could not be separated according to these criteria. In RS, two-thirds of cases were men, whereas the sex distribution in PSP-P was even. Disease duration in RS was significantly shorter (5.9 versus 9.1 years, P < 0.001) and age at death earlier (72.1 versus 75.5 years, P = 0.01) than in PSP-P. The isoform composition of insoluble tangle-tau isolated from the basal pons also differed significantly. In RS, the mean four-repeat:three-repeat tau ratio was 2.84 and in PSP-P it was 1.63 (P < 0.003). The effect of the H1,H1 PSP susceptibility genotype appeared stronger in RS than in PSP-P (odds ratio 13.2 versus 4.5). The difference in genotype frequencies between the clinical subgroups was not significant. There were no differences in apolipoprotein E genotypes. The classic clinical description of PSP, which includes supranuclear gaze palsy, early falls and dementia, does not adequately describe one-third of cases in this series of pathologically confirmed cases. We propose that PSP-P represents a second discrete clinical phenotype that needs to be clinically distinguished from classical PSP (RS). The different tau isoform deposition in the basal pons suggests that this may ultimately prove to be a discrete nosological entity.

Change in rates of cerebral atrophy over time in early-onset Alzheimer's disease: longitudinal MRI study.

The extent to which cerebral atrophy in Alzheimer's disease changes with time is unknown. We used multiple MRI scans to measure progression of cerebral atrophy in 12 patients with Alzheimer's disease who were followed up from a presymptomatic stage through to moderately severe dementia. Analysis with hierarchical regression models with quadratic terms in time provided evidence of increasing yearly percentage losses in brain volume. At the time when patients were judged to have mild dementia (mini-mental state examination score MMSE=23), mean yearly loss of brain volume was 2.8% (95% CI 2.3-3.3), which rose by 0.32% per year (0.15-0.50). Our findings reinforce the need for early diagnosis and therapeutic intervention in Alzheimer's disease.

The underlying risk of death after myocardial infarction in the absence of treatment.

BACKGROUND: The underlying risk of death in the absence of treatment after a myocardial infarction (MI) is poorly documented. METHODS: Analysis of 23 published studies in which 14 211 patients were followed prospectively after MI; 6817 deaths were recorded. We restricted the analysis to studies in which follow-up was completed by 1980 to quantify the underlying risk in the absence of effective treatments. RESULTS: After a first MI, on average, 23% of patients died before reaching the hospital and another 13% died during hospital admission; these rates increased with age. After hospital discharge cardiovascular mortality was approximately 10% in the first year and 5% per year thereafter, rates that were unrelated to age or sex. The yearly death rate of 5% persisted indefinitely; after 15 years, cumulative cardiovascular mortality was 70%. After a subsequent MI, 33% of patients died before reaching the hospital, and 20% died in hospital. After discharge, cardiovascular mortality was approximately 20% in the first year and 10% per year thereafter, rates again unrelated to age and sex. Approximately a third of all heart disease deaths occurred minutes after the first MI, a sixth during the first hospitalization, and half after a subsequent MI, which could occur many years after the first. CONCLUSIONS: In persons with a history of MI, cardiovascular mortality in the absence of treatment is high-5% per year after a first MI and 10% per year after a subsequent MI, persisting for many years and probably for the rest of a person's life. The high mortality rate emphasizes the need to ensure that everyone who has had an MI, even years previously, receives effective preventive treatment.

Endometrial thickness as a test for endometrial cancer in women with postmenopausal vaginal bleeding.

OBJECTIVE: To assess the value of endometrial thickness measurement as a test for endometrial cancer in postmenopausal women with vaginal bleeding (symptomatic women). DATA SOURCES: We conducted a literature search using the MEDLINE database from 1991 to 1997, and the key words "vaginal ultrasonography" and "endometrial thickness measurement." The review was limited to original research reports written in English, concerning symptomatic women having vaginal ultrasonography before a diagnostic test and not receiving tamoxifen. STUDY SELECTION: A total of 48 studies were identified. A questionnaire was sent to the corresponding author of each paper requesting supplementary information. Data were included in our analysis if the corresponding author was able to supply information on the median endometrial thickness in unaffected symptomatic women and the endometrial thickness values in affected women. Nine studies were thus included in our meta-analysis, representing 3483 women without endometrial cancer and 330 women with endometrial cancer. TABULATION, INTEGRATION, AND RESULTS: The median endometrial thickness in women with endometrial cancer was 3.7 times that in unaffected women at the same center, and with the same menopausal status and same hormone replacement therapy use category. The detection rate was 63% (95% confidence interval 58, 69) for a 10% false-positive rate, or 96% (95% confidence interval 94, 98) for a 50% false-positive rate. CONCLUSION: Endometrial thickness measurement in symptomatic women does not reduce the need for invasive diagnostic testing because 4% of the endometrial cancers would still be missed with a false-positive rate as high as 50%.