Question | Answer |
Steps in the EBM process | Ask, Acquire, Appraise, Apply |
Framing the question: | PICO: Patient, Intervention, Comparison, Outcome (& Type of question/study) |
Hierarchy of Evidence | Systematic Rvw/meta; RCT; Cohort/Case-Ctrl; Case Series, case report; Letters, etc |
Where to search the lit: | PubMed; Cochrane Collabn; Medline (OVID); ACP Journal Club; Practice Guidelines |
Appraise: questions | Are the results of study valid? What are the results? Will the results help in caring for my pt? |
the study of the frequency, distribution and determinants of health related states or events in specified populations | Epidemiology |
2 assumptions of epidemiology | 1: Human disease does not occur at random; 2: Factors that cause & prevent dz can be identified |
Defn Frequency: | how many cases of the dz exist or occurred |
Defn distribution: | who is getting the dz, where & when is it occurring (person, place & time) |
Defn Factors/Determinants: | what causes the disease, what influences or modifies the disease (physical, biological, social, cultural, behavioral factors) |
Measures of dz frequency | Prevalence; incidence |
A fraction where the numerator is contained in the denominator (e.g., percentage); falls between 0 and 1 | Proportion |
Expression of the relationship btw a numerator & a denominator where the 2 usually are separate & distinct quantities, neither being included in the other; No restrictions on ranges or dimensions | Ratio |
Total number of individuals with the outcome of interest (i.e., disease, injury, health benefit) within a defined population at a particular point or period in time | Prevalence |
The number of persons with the disease or attribute at a specific point in time | Point Prevalence |
Number of persons with the dz at any time during defined period | Period Prevalence |
Number of new events or cases of an outcome of interest (i.e., dz, injury, health benefit) that develop in a population of individuals at risk during a specified interval of time | Incidence |
Cumulative Incidence (AKA: Relative Risk) | Proportion: outcome occurring during a certain period of time |
Incidence Rate (AKA: Incidence Density) | Ratio: outcome occurring during person-time at risk |
Assumptions for cumulative incidence: | All study participants included in the calc for CI = observed (at risk) for the entire period; CI for specified time period infers* that the risk of dz is constant throughout that time period |
Considered a more precise estimate of the impact of exposure in a population because it utilizes all the information about time at risk | Incidence density |
Able to determine changes and fluctuations in risk of disease over time | Incidence density |
Defn population at risk | Free from dz, but susceptible |
Prevalence pool influenced by: | Incidence rate of dz; Dz Detection; recovery rate (dz duration); fatality rate |
Methods & procedures for collecting, classifying, summarizing, & analyzing data & for making scientific inferences from such data | Statistics |
Descriptive Statistics | Use of graphic, tabular, or numerical techniques to describe the properties of the data |
2 types of categorical variable | nominal; ordinal |
Categorical variable: no intrinsic order: | Nominal (e.g., race, gender, sub-types of leukemia) |
Categorical variable: have some intrinsic order | Ordinal variable (e.g., pain severity: none, mild, moderate, severe; Education level) |
Measures of central tendency | Means, medians, modes |
Measures of variability | Range, standard deviation (& variance) |
Median vs mode | median: middle value in list; mode = most frequent value |
If variable is normally distributed: | mean, mode, median all approx the same value |
Problems with the mean | 1: outliers skew mean; 2: bimodal dist: mean not a good summary stat |
Measure of how spread out distribution is | variance |
Positive sq root of variance | std deviation |
Units for std deviation | same units as the observations |
In normal dist, there is a defined relationship between: | std deviation & percentile distribution of the observations |
Empirical rule: Std deviation (1,2,3): percentages | 68, 95, 99.7% |
A very large group of units about which scientific inferences are to be made = | Population |
Subset of units in the underlying population = | Sample |
Inferential Statistics | Use of procedures to make inferences from the data (i.e., what are the implications of or what conclusions can be drawn from the data?) |
Provides the actual numerical information used in making inferences about the population | Sample |
Sample mean vs popn mean | Not nec same as popn; usu close to popn mean; some samples have means very diff from popn means |
P-value = probability that … | See a difference at least this large if samples were drawn from popn with same mean |
Defn Study Power | Probability that a study will conclude there is an association btw exposure & outcome when one truly exists |
Defn Study Power = ability of a study to: | to detect a statistically sig diff btw groups if one truly exists; 1 - beta |
Confidence intervals can be calculated around: | Diffs in means; diffs in proportions; odds ratios, relative risks, hazard ratios |
Confidence interval is presented as: | 1: point estimate; 2: plausible range of diffs btw which the true diff might actually lie (with 95% confidence) |
Mann-Whitney U-Test / Wilcoxon Rank Sum test: used when? | Used to compare continuous variables; Unpaired data; Nonparametric; Tests similar hypothesis as t-test |
Wilcoxon Signed Rank Test used for what data? | Used for paired data (e.g., before and after measurements on same individual) |
Wilcoxon Signed Rank Test = | Non-parametric test; Tests similar hypothesis as paired t-test |
Statistical test used to determine if there is an association between the row and column classifications | Chi squared test |
Assumptions of t test | Normally distribd variable (bell curve); independent observations (diff people, not multiple measurements on same people); reasonably large sample size |
t-test used to compare means of >2 groups? | No (2 groups only; otherwise: other tests) |
Purpose of t test | to detn how likely observed diffs in means btw groups could be due to chance |
Type I error | False positive (alpha error) |
Type II error | False negative (beta error) |
t-statistic corresponds to: | a given p-value |
The bigger the t-statistic: | the smaller the p value (more stat sig) |
Causes of larger t-statistic | Larger diff btw means; smaller variance; larger sample sizes |
Used to compare observations that are not independent | paired t-test |
Alternatives to t-test or paired t-test | Nonparametric tests |
Nonparametric tests are used when: | assumptions of t-test, particularly normal dist, are not met |
Nonparametric tests most often used when: | with small sample sizes |
Power: Nonparametric tests vs parametric tests | Nonparametric less powerful than parametric |
Categorical data are often presented in: | frequency tables |
Test: compares observed frequencies to expected frequencies | Chi squared test |
Larger chi-square statistic correlates to: | smaller p-value |
Variations on chi square test | Fisher exact; Mantel-Haenszel |
Fishers exact test used when: | expected frequencies in cells are small (<5) |
Mantel-Haenszel chi-square tests for: | linear association btw the row & column variables |
When interpreting results of a study, consider: | both the p-value and the size of the effect |
Alternative to p-values as a way of presenting and interpreting results | confidence intervals |
CI: single value most likely to represent the true diff btw groups = | point estimate |
CI provides info on: | precision of the estimate; statistical significance |
If 95% CI excludes the null value, then: | the findings are significant at p<0.05 |
CI: Greater variance results in: | wider confidence interval |
CI: Smaller sample size results in: | wider confidence interval |
Type I error: | Concluding that a difference exists btw groups when no difference exists |
Type II error: | Concluding that no difference exists btw groups when a significant difference does exist |
Alpha = | p-value |
Study Power is intimately related to: | the sample size used in a study |
Most studies designed with what power? | 80 or 90% power |
1-tailed vs 2-tailed test: which more conservative? | 2-tailed |
1-tailed vs 2-tailed test: which is the std in most research? | 2-tailed |
Chi squared used on what data type? | Categorical |
AKA Prevalence study | Cross-sectional study |
A study in which conditions are under the direct control of the investigator (e.g., randomized clinical trial, intervention study) | Experimental |
A study that does not involve an intervention or clinical trial. Investigators observe without intervening other than to record, classify, count and statistically analyze what is observed. | Observational |
______ studies examine the relationship between exposure and outcome with regard to the exposure preceding the outcome (“causation”) | Etiologic studies |
______ studies describe either the exposure, the outcome, or both, but do not infer a causal relationship between the two (association) | Descriptive studies |
A careful & detailed report of a single pt by one or more HCPs; documents unusual medical occurrences; possibly first clues in identification of new dz | Case Report |
An expansion of a case-report which usually describes the characteristics of a number of pts; may occur within a short period of time | Case series |
Study in which the units of analysis are populations or groups (not individuals); a study of group characteristics | Ecologic study |
A study that examines the relationship between diseases and other variables of interest as they exist in a defined population at one particular time. | Cross-sectional study |
Snap shot of disease prevalence in a population at one point in time | Cross-sectional study |
Can a cross-sectional study determine a temporal relationship (i.e., exposure preceded outcome)? | No |
Observational study: subjects selected on basis of whether they have (cases) or do not have (controls) a particular dz. Groups are compared w/respect to the proportion having a history of the exposure of interest. | Case-control study |
Source populations for cases in case-control studies | Hospitals (Inpatient; Outpatient/Clinics); Communities (Registries; Dz Surveillance Registries) |
Case-control studies: 2 types of matching of cases to controls | Frequency matching; individual matching |
Case-control studies: Number of controls selected per case can be influenced by: | Cost; recruitment; small number of cases |
Case-control studies: There is only a small increase in statistical power when the (controls:cases) ratio increases beyond: | 3:1 |
Differences in the way exposure information is remembered or reported by cases (who have experienced an adverse health outcome) and controls (who have not). | Recall bias |
2 types of odds ratio | Prevalence odds ratio; incidence odds ratio |
Case-control studies: diagnostic criteria should minimize the likelihood that: | 1: true cases are missed (sensitive); 2: false cases are included (specific) |
A hybrid design in which a case-control study is nested in a cohort study. | Nested case-control study |
Observational study: subjects are sampled based on presence (exposed) or absence (non-exposed) of a risk factor of interest; subjects followed over time, assess devt of dz/ outcome of interest | Cohort studies |
Type of cohort study: investigators use historical data to reconstruct a timeline of exposure | Retrospective cohort |
Type of cohort study: investigators identify the original population at the beginning of the study and follow them concurrently over time | Prospective cohort |
Cohort study: fixed cohort | Exposure groups in a cohort study represent groups that are defined at the start of the follow-up or observation period. No new members are added |
Cohort study: dynamic cohort | Study cohort gains members over time through in-migration, or loses members through out-migration |
Best observational design to establish cause and effect relationships = | cohort study |
The ratio of the risk of disease among the exposed to the risk among the unexposed | Relative Risk (AKA: Cumulative Incidence Ratio, Risk Ratio) |
The ratio of the rate of disease in the exposed population to the rate in the unexposed population | Rate Ratio (AKA: Incidence Density Ratio) |
Ecologic study: Advantages | Quick; Inexpensive; Hypothesis generating; Useful for establishing associations at an environmental level |
Ecologic study: Disadvantages | Individual exposure information is not collected; so we don’t know if those who are exposed are actually developing the dz and vice-versa |
Descriptive studies: Advantages | Inexpensive; Quick; Hypothesis generating; Assessment of public health interventions |
Descriptive studies: Disadvantages | Temporal relationship cannot be determined; Incidence (risk) of disease cannot be determined; Causal link cannot be established |
Study: subjects randomly allocated to receive new active treatment, or the std tx, or a placebo. Different tx groups are then compared to assess effectiveness of the treatment | Randomized clinical trial |
Types of RCT | Parallel, Equivalency (parallel), Crossover, Factorial, Group allocation |
RCT: ____-blind trial = good for trials on very sensitive topics or subjectively measured outcomes | Triple-blind |
Randomization in RCT | Random allocation to a grp; group in which the subject is placed is random, not the subject himself; core of a RCT |
Random Selection in RCT | Selecting subjects, at random, to be part of a study population; can be conducted in RCTs, but primarily in intervention and observational studies |
RCT: types of randomization | Simple, blocked, stratified |
RCT: blocked randomization requires: | a double-blind RCT design to eliminate investigator bias |
RCT: Only disadvantage of blocked design: | analyses become more cumbersome |
If blocked randomization is not double-blinded, then: | Block size pattern can randomly vary within a study (blocksize = 2, 4, 6, or 8) |
Efficacy vs Effectiveness | Efficacy = can it work (under lab conditions)? Effectiveness = will it work (are results generalizable to larger population)? |
RCTs vs Observational study: RCTs are good for: | Establishing tx efficacy; understanding effectiveness of intervention when small sample size; understanding: 1- efficacy for tx for serious/ rare dz; 2-dz pathology in a unique population |
RCTs vs Observational study: Observational study good for: | Understanding associations in large pops; establishing need to conduct an RCT; understanding associations of prognostic factors with dz of interest prior to RCT |
Sensitivity = | True Positives / (True Pos + False Negs) |
Sensitivity will be higher if there are: | fewer false negatives; i.e., fewer Type II errors |
Specificity = | True Negatives / (True Negs + False Pos) |
Specificity will be higher if there are: | fewer false positives; i.e., fewer Type I errors |
Positive Predictive Value (PPV) = | True Pos / (True Pos + False Pos) |
Negative Predictive Value (PPV) = | True Negs / (True Neg + False Neg) |