R
Filomena Pereira-Maxwell in Medical Statistics, 2018
The magnitude of the risk ratio speaks to the strength of the association between disease and exposure, or treatment and outcome (the statistical significance, however, does not). It will, however, depend on length of follow-up, so that if measured at the end of a long period of follow-up for an outcome such as ‘death’ it will tend toward the null value of 1 (HENNEKENS, BURING & MAYRENT (eds.), 1987). Rate or hazard ratios may give a better estimate of relative risk in these situations. Also, RR cannot be larger than 1/’risk among the unexposed’ (or smaller than ‘risk among the treated’/1), so that if the risk in these groups is high (common events) the risk and rate ratios will differ more so (ROTHMAN, 2012). For rare diseases, risk, rate and odds ratio (OR) will be approximately equal (KIRKWOOD & STERNE, 2003). In cumulative case-control studies (if the disease is rare) and case-cohort studies, the odds ratio gives an estimate of the risk ratio. The attributable fraction and the relative risk reduction (depending on whether the exposure is harmful or protective) are measures ofrelativeimpact that may be calculated from the risk ratio (or from the absolute risk difference, which is a measure of net impact). See also measures of effect, measures of association.
Exact Likelihood Methods for Group-Based Summaries
Christopher H. Schmid, Theo Stijnen, Ian R. White in Handbook of Meta-Analysis, 2020
In this section, we focus on the situation where two groups are compared with respect to the probability of some outcome event of interest. We suppose that the data for each of I studies consist of a 2 × 2 table (numbers of subjects with/without event for two groups). The most popular choice for the effect measure is the (log) odds ratio and we will use that choice. Other effect measures such as the risk ratio or the risk difference lead to models that are very difficult to fit. The number of events in a group follows a binomial distribution and in Section 5.3.1 we will use the binomial likelihood in order to avoid the normal within-study distribution assumption of Chapter 4. All problems with the two-stage approach mentioned in Table 5.1 apply for dichotomous outcomes, including number 3, which did not apply for the continuous outcome case discussed in the previous section. These problems get worse as the numbers of events get smaller.
Search for causes of disease occurrence: Why does disease occur?
Milos Jenicek in Foundations of Evidence-Based Medicine, 2019
Two aspects of specificity34 should be studied. The manifestational specificity reveals that a factor (cause under study) leads to a consistent pattern of consequences. In other terms, clinical spectrum and gradient of morbid manifestations (one or more diseases) should be exclusive to a given exposure. Cluster analysis and analysis of categorical data may be helpful in the study of manifestational specificity. For example, the ‘four D's’ of pellagra (dermatitis, diarrhea, dementia, death) are specific to this kind of malnutrition. The causal specificity is based on a measure of how important one factor is relative to others. This criterion may be evaluated by assessing differences between rates of occurrence in exposed and unexposed subjects, in absolute terms of attributable risk (syn. risk difference), or in relative terms as etiological fraction (syn. attributable risk percent, attributable fraction). Absolute specificity means that all cases are due to one factor (100% of cases are attributable to the factor under study). The less specific the factor (and the causal association linked to it), the lesser the proportion of disease cases are due to exposure. This is discussed in detail in Section 8.4. NB Readers of specific literature should not be surprised that the summation of all known etiological fractions of various factors, as provided by different studies, can yield more than 100% of the total etiology. In each original study, the combination of the factors under study and all other etiological factors (minimal causes) can be different. The information on etiological fraction of a causal factor is primarily valid within the study. A proper integration of information on etiological fractions coming from various studies is still being developed.
Homogeneity test of relative risk ratios for stratified bilateral data under different algorithms
Published in Journal of Applied Statistics, 2023
Ke-Yi Mou, Chang-Xing Ma, Zhi-Ming Li
In medical clinical studies, observations from patients' paired parts (e.g. eyes, ears, and arms) are usually collected as paired data. The paired outcomes for each patient will be no, unilateral or bilateral response(s). Data from all patients can be summarized in a contingency table. The correlation between responses from paired parts should be taken into account to avoid biased or misleading results. In clinical practice, research subjects often can be distinguished by some control variables (e.g. age, gender), which contribute to stratified data. Although the questions involving treatment-by-stratum interaction are often secondary in most multi-center trials, they are still important as preparatory work for the overall and subgroup analyses. For a stratified bilateral design with two groups, the interaction can be tested by comparing different ratios across strata. If the ratios are not significantly different, the effect of stratum is negligible. Relative risk ratio, odds ratio and risk difference are often used to quantify the strength of the association. Generally, relative risk ratio is more visual than odds ratio. Walter [22] pointed out the population risks of some diseases were rather small such that risk differences between groups were less dramatic. Thus, risk relative ratio can be effectively used to study the homogeneity test in stratified bilateral data.
Control of Severe, Life-Threatening External Bleeding in the Out-of-Hospital Setting: A Systematic Review
Published in Prehospital Emergency Care, 2021
Nathan P. Charlton, Janel M. Swain, Jan L. Brozek, Maja Ludwikowska, Eunice Singletary, David Zideman, Jonathan Epstein, Andrea Darzi, Anna Bak, Samer Karam, Zbigniew Les, Jestin N. Carlson, Eddy Lang, Robby Nieuwlaat
GRADEPro software (Guideline Development Tool [Software], McMaster University, gradepro.org) was used to create Evidence Profiles (EPs) for reporting summary of findings and certainty of the evidence per comparison and outcome. Dichotomous outcomes were reported as relative risk (RR), odds ratio (OR), or risk difference (RD). If a measure of association was not reported, the authors calculated an unadjusted RR for the study using MedCalc for Windows Version 19.07 (MedCalc Software, Ostend, Belgium). When presented in the primary study, adjusted ORs (aOR) and adjusted RRs (aRR) were also reported. Continuous outcomes were reported as mean difference (MD), and standardized mean difference (SMD), if available. For single-arm studies, we reported results as published (means or proportions) and provided the median among all included studies, or in some studies, when limited data was available, reported minimum and maximum.
Epidemiology and prevention of oesophageal adenocarcinoma
Published in Scandinavian Journal of Gastroenterology, 2022
In the meta-analysis, the risk of GORD was slightly increased in women compared to men (odds ratio (OR) 1.18, 95% CI 1.15–1.20), but there was no clear association with age (Table 1) [5]. The risk difference by sex is small and conflicting between studies and possible reasons for this difference is unclear. The risk of GORD increased with increasing body mass index (BMI) (OR 1.73, 95% CI 1.58–1.89 for BMI ≥30 compared to BMI 18.5–29.9), while tobacco smoking was not associated with GORD. Low socioeconomic status, represented by a low educational level was also associated with increased risk of GORD (OR 2.11, 95% CI 1.99–2.24), compared to medium and high educational level. Use of non-steroid anti-inflammatory drugs or aspirin was also associated with increased risk of GORD (OR 1.46, 95% CI 1.33–1.60).
Related Knowledge Centers
- Colorectal Cancer
- Number Needed to Harm
- Relative Risk Reduction
- Incidence
- Number Needed to Treat
- Number Needed to Harm
- Standard Score
- Statistical Significance
- Population Impact Measure