Epi 546/547 Review

Absolute Risk Reduction (ARR, aka Attributable Risk): The difference in risks of an outcome between two experimental groups, or the additional risk of disease following exposure over and above that experienced by people not exposed (if one group is a control). ARR = CER - EER or ARR = EER1 - EER2 Accuracy: truthfulness of results or measurements, also referred to as validity. Bayes Theorem: A unifying methodology to calculate the probability of disease given either a positive or negative test result for any combination of Se, Sp, and Prevalence. Bias: Systematic error in study design which may skew results leading to a deviation from the truth. There are three broad classes of bias: Confounding, Selection, and Measurement. Interviewer Bias: Error due to an interviewers conscious or subconscious gathering of selective data Lead-Time Bias: Mistakenly attributing increased survival of patients to a screening intervention when longer survival is only a reflection of earlier detection in the preclinical phase of the disease. Measurement Bias: Occurs when one group of patients has a better (or worse) chance of having their outcome detected than another group. More likely to occur for soft outcomes that are less clear cut. Can minimize by adhering to the following principles: ensure that all observations are carried out by observers who are blinded to the exposure status of the particular patient, develop (and use) careful criteria or rules for deciding whether an outcome event has occurred, and apply equally rigorous efforts to ascertain all events regardless of exposure group. Recall Bias: Error due to differences in accuracy or completeness of recall to memory of past events or experiences. Particularly relevant in case control studies (CCS). Referral Bias: The proportion of more severe or unusual cases tends to be artificially higher at tertiary care centers. Selection Bias: An error in patient assignment between groups that permits a confounding variable to arise from the study design rather than by chance alone. Assembly Bias: when the groups of exposed and non-exposed patients assembled for the study differ in some way other than the prognostic factors under study (aka susceptibility bias). Migration Bias: When patients drop out of a study prematurely (aka LTFU). If drop outs occur randomly then no bias occurs. But usually subjects have a reason to drop out and most of those factors are related to prognosis (death, recovery, SE of treatment, disinterest in study). If it occurs on a large scale, exposure groups become less comparable as time progresses. Less than 5% LTFU is considered inconsequential, >20% is considered serious. Best/Worst Case Analysis: Used to estimate how large migration bias is. The number of subjects LTFU are firs assumed to have not developed the outcome of interest (best case scenario), then are assumed to have all developed the outcome (worst case scenario). Comparing the final results under these two scenarios gives the possible effects of migration bias. Generalizability Bias: Selective referral of patients to tertiary medical centers - they are highly selected and may be very different in terms of age, severity of disease, presence of complications from those seen in other environments (primary care clinics). Influences the natural history of the condition under study. Spectrum Bias: Occurs because of a difference in the spectrum and severity of disease between the population where the diagnostic test was developed and the clinical population that the test is applied to. The disease subjects in the development population tend to be the sickest of the sick with few FN results (so sensitivity is overestimated), while the non-disease population tends to be the wellest of the well with few FP results (so specificity is overestimated). Net effect is that the initial Se and Sp values reported for a new diagnostic test are often overly optimistic and it is only when the test is applied to reallife clinical populations that it is found to be not so nearly useful as the first reports indicated. This is a form of selection bias.

Epi 546/547 Review

Verification Bias: The index test itself is used as a criterion to select which patients receive the definitive (gold or referent standard) diagnostic procedure which is either invasive and/or expensive. Results in overly optimistic estimate of sensitivity and an underestimate of specificity (aka work-up bias or test-referral bias). The tests performance can then be biased because patients with negative test results are less likely to be evaluated with the traditional gold standard. Volunteer Bias: people who choose to enroll in clinical research may be systematically different (healthier, more motivated) than your patients. Blinded/Masked: Blinded studies purposely deny access to information in order to keep that information from influencing some measurement, observation, or process (reduces information bias). Double-blinded refers to the fact that neither the study subject not the study staff are aware of which group or intervention the subject has been assigned. Ideally everyone who is blinded or not should be explicitly identified. Prevents measurement bias, improves compliance, helps retain study participants, and prevents cointerventions. Co-Intervention: Interventions other than the treatment under study. Particularly relevant to therapy RCTs - readers should assess whether co-interventions were differentially applied to the treatment and control. Concealment: A fine point associated with randomization that is very important. Ideally, you want to be reassured that the randomization schedule of patients was concealed from the clinicians who entered patients into the trial. Thus the clinician will be unaware of which treatment the next patient will receive and therefore cannot consciously - or subconsciously - distort the balance between the groups. If randomization wasnt concealed, patients with better prognoses may tend to be preferentially enrolled in the active treatment arm resulting in exaggeration of the apparent benefit of therapy (or even falsely concluding that the treatment is efficacious). Concealment reduces the possibility of selection bias. Allocation Concealment: process by which study participants and clinicians are kept unaware of the upcoming assignment. Prevents selection bias. Studies should describe what concealment measures were taken - sequentially numbered opaque sealed envelopes, use of off-site randomization center, secure computer-assisted methods, etc. Confidence Interval: Clinical research provides a point estimate of effect from a sample of patients; CIs express the degree of uncertainty or imprecision regarding this point estimate. It represents a range of values consistent with the experimental data. CIs and its associated point estimate help us make inferences about the underlying population. The commonly used CI can be defined as the range of values within which we can be 95% sure that the true underlying value lies. It estimates the sampling variation by adding and subtracting 2 standard errors from the point estimate. It is affected by inherent variability of the characteristic being measured, and the study sample size - the larger the sample size the narrower (and more precise) the CI. Values outside of a 95% CI are statistically significantly different (at P<0.05) from the point estimate. Confounder/Confounding Variable: A factor that distorts the true relationship of the study variable of interest by virtue of being related to both the study variable and the outcome of interest. Confounders are often unequally distributed among the groups being compared. Randomized studies are less likely to have their results distorted by confounders because randomization should result in the equal balance of these factors at baseline. Cox Regression Model: A regression technique for survival analysis that allows adjustment for known differences in baseline characteristics between intervention and control groups when applied to survival data. Diagnosis: The determination of the nature of a disease; a process of more or less accurate guessing. Effectiveness: A measurement of benefit resulting from an intervention for a given health problem under conditions of usual practice. This form of evaluation considers both the efficacy of an intervention and its acceptance by those to whom it is offered. It helps answer does the practice do more good than harm to people to whom it is offered?

Epi 546/547 Review

Efficacy: A measure of benefit resulting from an intervention for a given health problem under conditions of ideal practice. Helps answer does the practice do more good than harm to people who fully comply with the recommendations? Event Rate (risk or CIR): The risk or CIR of an event. Calculated as the proportion of a fixed population who develop the event of interest over a period of time. In a RCT design the terms CER and EER refer to the risks in the two comparison groups. Generalizability (or External Validity): The extent to which the conclusions derived from a trial/study can be used beyond the setting of the trial and the particular people studied in the trial. Are the results applicable to the full population of all patients with this condition? Harm: Unexpected or expected negative consequences of treatments/interventions. The tolerance for harmful effects at the patient/societal/medical-legal level is set very low. Identification is difficult for the following reasons: the determination of cause and effect has to rely on observational studies (Cohort or CCS), most purported serious side effects are rare, and the underlying reason for prescribing the drug may act as a confounder of the association between the drug and the side effect. Confounding in these studies can result from two different mechanisms: Confounding by indication: the underlying reason for prescribing a drug may act as a confounder of the associated between the drug and the adverse outcome. Channeling effect: tendency for clinicians to prescribe certain treatments based on a patients underlying prognosis or comorbidity profile, resulting in differences in baseline risk and potentially biased treatment effects. Studies should statistically adjust for these baseline differences in risk using multivariable logistic regression models. Only absolute fix for these problems is to do an RCT. Hazard Ratio: The relative risk of an outcome over the entire study period; often reported in the context of survival analysis (Cox regression model). Has a similar interpretation to the relative risk. Heterogeneity: Differences between patients (clinical heterogeneity) or differences in the results of different studies (statistical heterogeneity). Sources include population, intervention, outcome, design. Q statistic: A statistical test for heterogeneity based on the chi-square test. The null hypothesis is that there is homogeneity among the results (i.e., p>0.05), and if p<0.05, then heterogeneity is present. I2 statistic: quantitative measure of heterogeneity. <25% = low heterogeneity, 25-75% = moderate heterogeneity, >75% = high heterogeneity. Pooling: Methods include a weighted average (larger trials have more weight because a simple mean may provide an unbalanced estimate of the effect size), fixed-effects model, random-effects model. Fixed-effects model: Inference based on the studies at hand, accounts only for within-study random variation. Random-effects model: Random sample of studies from all the possible studies in the universe, accounts for between-study and within-study random variation. Here, more weight is given to smaller studies, and have a wider 95% CI. Inception Cohort: A designated group of persons assembled at a common time early in the development of a specific clinical disorder and who are followed thereafter. In assessing articles about prognosis it is critical that the inception cohort is well described in order to permit assessment of the homogeneity of the cohort. Ideal point of inception is near the onset of the disease, but regardless of the starting point used, it is critical that it is the same for all subjects in the study. If a survival cohort is used instead of a true inception cohort, assembly bias can occur. Incidence Rate: Number of new cases of disease occurring during a specified period of time; expressed either as a percentage (or proportion) of the number of people at risk (CIR) or the number of new cases occurring per person time (IDR). ITT Analysis: Analyzing patient outcomes based on which group they were randomized into regardless of whether they actually received the planned intervention. This analysis preserves the power of randomization, thus maintaining that important unknown factors that influence outcome are likely equally distributed in each comparison group. It is the most conservative but valid analytical approach for an RCT (compared to

Epi 546/547 Review

as treated or per protocol analysis). A modified ITT is an analysis where the investigators excluded a small number of subjects from the pure ITT population (ex: patients who should not have been enrolled in the study or those who died shortly after enrollment of unrelated causes). Internal Validity: The degree to which inferences drawn from a specific study are accurate. It requires a careful assessment of the studys methodology to determine whether the observed findings are accurate. It implies that apart from random error the studys findings cannot be ascribed to a systematic error or bias; in other words the study does not suffer from confounding, selection or information bias to an important degree (judgment is required here). RCTs have high internal validity. Kappa: A measure of reliability or agreement between two raters for categorical or qualitative data (e.g., two physicians independently reading x-ray films). Kappa adjusts for the agreement that would be expected to occur due to chance alone, and is thus referred to as a chance-corrected agreement. Kappa is preferred over other agreement measures such as the overall % agreement which is highly influenced by the prevalence of the condition being evaluated. It ranges from -1 to +1. Values above 0.8 indicate excellent agreement, values 0.6-0.8 indicate substantial agreement, vales 0.4-0.6 indicate moderate agreement, and values <0.4 indicate fair or poor agreement. Likelihood Ratio: A ratio of likelihoods (or probabilities) for a given test result. The probability of observing the test result in the presence of disease divided by the probability of observing the test result in the absence of disease. It is the odds that a given test result would occur in a diseased individual compared to a non-disease individual. LR+ = Se/1-Sp = TPrate/FPrate = (A/A+C)/(B/B+D) LR- = 1-Se/Sp = FNrate/TNrate = (C/A+C)/(D/B+D) Advantages: using the odds-likelihood ration form of Bayes theorem we can easily adjust the test characteristics for the underlying prevalence of disease. Shows that the environment in which the test is applied is as important as the information provided by the test. Also, can be calculated for a range of test results, preserving clinical information and provides a succinct illustration of how clinicians can work more efficiently. Pre-test odds x LR = Post-test odds Pre-test odds = Prev/1-Prev Post-test probability = Post-test odds/1+Post-test odds Disadvantage: most people have difficulty in converting probabilities to odds (and vice versa). Can use the Fagan nomogram for conversion instead of good old math. Matching: A deliberate process to make the study group and comparison group comparable with respect to factors (or confounders) that are extraneous to the purpose of the investigation but which might interfere with the interpretation of the studies findings. For example, in CCS, individual cases may be matched with specific controls on the basis of comparable age, gender, and/or other clinical features. Meta-Analysis: A systematic review which uses quantitative tools to summarize the results. In systematic reviews, each study is a subject. Make sure they performed a comprehensive literature search - no language restriction, searched all available databases (embase, medline, cochrane central), and the grey literature (experts, conference proceedings, unpublished). Should have a flow chart of included/excluded studies. Multivariable Regression Analysis: A type of regression model that attempts to explain or predict the dependent variable (or outcome variable or target variable) by simultaneously considering 2 or more independent variables (or predictor variables). Used to account for confounding and interaction effects. Examples include multivariable logistic regression (for binary outcomes) and multivariable linear regression (for continuous outcomes). Non-Inferiority Trials: Trials undertaken with the specific purpose of proving that one treatment is no worse than another treatment, which is usually the current standard of care. Number Needed to Harm (NNH): The number of patients who would need to be treated over a specific period of time before one adverse side-effect of the treatment will occur. NNH = 1/ARI = 1/(CER)x(RR-1)

Epi 546/547 Review

Number Needed to Treat (NNT): The number of patients who need to be treated over a specific period of time to prevent one bad outcome. It is important to specify the treatment, its duration, and the bad outcome being prevented. NNT = 1/ARR = 1/(CER)x(1-RR) Odds: A ratio of probability of occurrence to non-occurrence of an event. O = P/1-P Odds Ratio: Most often used in CCS designs to describe the magnitude or strength of an associated between an exposure and the outcome of interest. Because the actual underlying disease risks (or CIRs) in the exposed and unexposed groups cannot be calculated in a CCS design, the OR is used to approximate the RR. It more closely approximates the RR when the outcome of interest in infrequent or rare (<10%). As a measure of the strength of association between an exposure and outcome the OR has the same interpretation as the RR. Calculated as the cross-product ratio. OR = AxD/BxC P-value: The probability of obtaining the value of the test statistic at least as large as the one observed, under the assumption that the null hypothesis is true. The smaller the P-value, the lower your degree of belief is in the null being true. It is used to make a decision based on the available data. Population Attributable Risk (PAR): The incidence of disease in a population that is associated with a risk factor. Calculated from the attributable risk (ARR) and the prevalence of the risk factor in the population. PAR = ARR x prevalence Population Attributable Risk Fraction (PARF): The fraction of disease in a population that is attributed to exposure to a risk factor. Under the assumption that the risk factor is a cause of the disease, it represents the maximum potential impact on disease incidence if the risk factor was removed. Calculated from the PAR. PARF = PAR/total disease incidence or PARF = P(RR-1)/[1 + P(RR-1)] Power: Ability to detect a difference between two experimental groups if one in fact exists (1-beta). Precision: A measure of variability in the point estimate as quantified by the confidence interval. Influenced by random error. Predictive Value: Prevalence has an influence on these measures - the interpretation of test results depends on the probability of disease before the test was run (prior probability). Positive Predictive Value (PVP): Proportion of people with a positive test who have the disease. Negative Predictive Value (PVN): Proportion of people with a negative test who are free of disease.

Epi 546/547 Review

Prevalence: Proportion of persons affected with a particular disease at a specified time. Prevalence rates obtained from high quality studies can inform clinicians efforts to set anchoring pretest probabilities for their patients. In diagnostic studies, also referred to as prior probability. Prognosis: The prediction of the future course of events following the onset of disease. Can count a full range of manifestations that may be important to the patient (death, complications, morbidity, disability, social/occupational function). Prognostic Factors: Factors that are associated with a particular outcome (such as death) among subjects that have disease. Examples are age, co-morbidities, tumor size, severity of disease, etc. Natural History: The evolution of the disease without medical intervention. Describes prognosis of patients if nothing is done for their ailment. Can only be studied here if unrecognized (asymptomatic anemia, pre-htn, occult cancers) or a normal discomfort (arthritis, mild depression, low back pain, etc). Natural history studies permit the development of rational strategies for attempting the early detection of untoward consequences of the disease and treatment of various disease. Studying this usually requires a cohort study, typically retrospective. One must ensure that the outcome in question has bee sought with the same intensity in both groups being compared (to prevent surveillance/detection bias) Clinical Course: evolution of disease in response to medical intervention. Commonly used measures: 5 year survival rate - CIR that measures the proportion of the original patient population that is alive after 5 years. These are easy to interpret and remember, but they fail to indicate the mortality rate at any point in time during the 5 year period. Case Fatality Rate - a type of CIR that measures the risk of death among those individuals with the disease. Must be accompanied by a specific time interval to have a meaningful interpretation. Mortality/Death Rate - IDR of death per population time. Measures speed of death due to a specific disease in a population-at-risk. Censoring: When the event of interest does not occur in all individuals in a prognostic study because: the study was stopped before everyone had the event, LTFU, or death from other causes. Survival Function: the probability of survival to a given point in time - plotted on a survival curve Hazard Function: the probability of an event at a specific moment in time (t), given the patient has already survived to that point in time. Is closely linked to the survival function, indicates the probability of the patient failing during the next time period. Kaplan-Meier Estimator: A widely accepted method of estimating the survival function. Also plotted on a curve-type graph. Randomization: Allocation of individuals to groups by a formal chance process such that each patient has an independent, equal chance of selection for the intervention group. Limits confounding bias. Relative Risk: The relative probability or risk of an event or outcome in one group compared to another. In a RCT, it is a measure of the efficacy (or magnitude) of the treatment effect. In a cohort study, where it is similarly used to express the magnitude or strength of an association, it is calculated as the ratio of risk of disease or death among an exposed population to the risk among the unexposed population. RR = EER/CER Relative Risk Reduction: The percent reduction in an outcome event in the experimental group as compared to the control group. It is the complement of RR or the proportion of risk removed by the intervention. Unlike the ARR, the RRR is assumed to be a constant entity - this is, it is assumed not to change from one population to another. It therefore represents a fixed measure of the efficacy of an intervention. RRR = 1 - RR or RRR = CER - EER/CER x 100 or RRR = ARR/CER Reliability: Refers to consistency or reproducibility of data. Referred to as agreement when examining categorical data. Intra-rater reliability refers to the consistency within the same observer/instrument. Interrater reliability refers to the consistency between two observers or instruments. ROC Curves: Receiver operator characteristic curves plot test sensitivity (on the Y axis) against 1 - Sp (on the X axis) for various cut-points of a continuously distributed diagnostic variable. The curves describe the tradeoff between Se and Sp as the cut point is changed. Tests that discriminate well crowd towards the upper north-west corner of the graph. ROC curves can be used to compare the discriminating ability of two or more tests by comparing the area under the curve (AUROC).

Epi 546/547 Review

Sensitivity: The proportion of people with disease who have a positive test TP/TP+FP SnNout: When a test with a high sensitivity is negative, it effectively rules out the diagnosis of disease. If the test is negative you can be confident that disease is absent, because FN results are rare. Sensitivity Analysis: A test of the stability of conclusions by evaluating the outcome over a range of plausible estimates, value judgments, or assumptions. Specificity: The proportion of people without disease who have a negative test TN/TN+FN SpPin: When a test is highly specific, a positive result can rule in the diagnosis. If the test is positive you can be confident that disease is present, because FP results are rare. Study Designs:

Epi 546/547 Review

Case Series: A collection or a report of the series of patients with an outcome of interest. No control group is involved. Case Control Study (CCS): Identifies patients who have a condition or outcome of interest (cases) and patients who do not have the condition or outcome (controls). The frequency that subjects are exposed to a risk factor of interest is then compared between the cases and controls. Because of the design of the CCS, disease rates cannot be directly measured. Thus the comparison between cases and controls is actually done by calculating the odds of exposure in cases and controls. The ratio of these 2 odds results in the OR, which is usually a good approximation of the RR. CCSs can investigate harm. Advantages: it is relatively quick and inexpensive, requiring fewer subjects than other study designs. It is often times the only feasible method for investigating very rare disorders or when a long lag time exists between an exposure of interest and development of the outcome/disease of interest. It is also helpful in studies of outbreak investigations where a quick answer followed by a quick response is required. Disadvantages: recall/interviewer bias, unknown confounding variables, difficulty selecting appropriate controls (selection bias), time order (did exposure status change because of early signs of the adverse outcome), cant provide estimates of absolute risk. Crossover Design: A method of comparing two or more treatments or interventions in which all subjects are switched to the alternate treatment after completion of the first treatment. Typically allocation to the first treatment is by a random process. Since all subjects serve as their own controls, error variance is reduced. Cross-Sectional Survey: The observation of a defined population at a single point in time or during a specific time interval. Exposure and outcome are determined simultaneously. Also referred to as a prevalence survey because this is the only epidemiological frequency measure that can be measured. Cohort Study: Involves identification of two groups (cohorts) of patients who are defined according to whether they were exposed to a factor of interest. The cohorts are then followed over time and the incidence rates for the outcome of interest in each group are measured. The ratio of these incidence rates results in the RR which quantifies the magnitude of association between the factor and outcome (disease). When the follow-up occurs in a forward direction the study is referred to as a prospective cohort. When follow-up is done based on historical information it is referred to as a retrospective cohort. Cohort studies can investigate diagnosis and prognosis. Advantages: Can establish clear temporal relationships between exposure and disease onset. Are able to generate incidence rates. Disadvantages: Control groups may be difficult to identify, exposure to a variable may be linked to a hidden confounding variable, blinding is often not possible, randomization is not present. For relatively rare diseases of interest, they require huge sample sizes and long follow-up, thus they are slow and expensive. N-of-1 Trial: When an individual patient undergoes pairs of treatment periods organized so that one period involves use of the experimental treatment and the other involves use of a placebo or alternate therapy. Ideally the patient and physician are both blinded, and outcomes are measured, treatment periods are replicated until patient and clinician are convinced that the treatments are definitely different or definitely not different. Randomized Control Trial (RCT): A group of patients is randomized into an experimental group and into a controlled group. These groups are then followed up and various outcomes of interest are documented. RCTs are the ultimate standard by which new therapeutic maneuvers are judged. Randomization should result in the equal distribution of both known and unknown confounding variables into each group. An unbiased RCT also requires concealment and where feasible, blinding. Advantages: Prevents confounding bias, generates unpredictable random allocation sequence and the concealment of this sequence - preventing selection bias. If study is blinded, it also prevents measurement bias. We can confidently attribute cause and effect with a good RCT. Disadvantages: often impractical, limited generalizability, volunteer bias, significant expense, and sometimes ethical difficulties.

Epi 546/547 Review

Systematic Review: A formal review of a focused clinical question based on a comprehensive search strategy and structured critical appraisal designed to reduce the likelihood of bias. No quantitative summary is generated. Meta-Analysis: A systematic review which uses quantitative methods to combine the results of several studies into a pooled summary estimate. Survival Analysis: A statistical procedure used to compare the proportion of patients in each group who experience an outcome or endpoint at various time intervals over the duration of the study. Validity: Truthfulness or believability of study conclusions or the extent to which a test actually measures what it is supposed to measure or accomplishes what it is supposed to accomplish. Aka accuracy. Validity implies the presence of a gold standard to which data can be compared. Economic Analyses: Cost-Effectiveness Analysis: A method for comparing different treatment decisions or management practices by their relative costs and effectiveness. Cost-effectiveness = net cost/net benefit. Marginal cost-effectiveness ratio (MCER): an estimate of the cost per unit of effectiveness for two competing strategies. Use a diagram with cost on Y axis and effectiveness on X axis. Anything in upper left quadrant is dominated, anything in lower right is dominant. If B is in the LLQ or RUQ, than whether A should be chosen over B depends on a qualitative assessment of how much one is willing to pay for added effectiveness or how much less effectiveness one is willing to have in order to save money. In the US, a general rule is that interventions that cost about $50k per life year gained or less are generally considered worthwhile.

A limitation of CEA is that it provides no information about what interventions across a broad spectrum of diseases and conditions should be used. QALYs: Quality Adjusted Life Years, one method to compare outcomes across different diseases or conditions. Derived by adjusting the length of survival by the quality of life of a particular health state. Measured by utilities that range from 0 to 1. Cost-Benefit Analysis: An alternative to CEA, it evaluates alternative management strategies in which costs and benefits are both expressed in the same terms (i.e., dollars). Can compare the net benefit for each program by subtracting costs from benefits. It is seldom used because it requires outputs to be expressed in monetary terms, and this is difficult to do with health outcomes. Perspective: Gold standard is societal. Others include patient, doctor, hospital, payers. A societal perspective requires that the indirect costs associated with the time of patients and unpaid caregivers are included in all costs estimates. Costs: Direct costs: the value of all goods, services, and other resources that are consumed in the provision of an intervention including side effects and other current and future consequences linked to the intervention. Charges are often used as a surrogate for things like hospital costs, professional fees,

Epi 546/547 Review

diagnostic tests, drugs and supplies. Charges tend to inflate true costs d/t profit margins, market imperfections, administrative inefficiencies, cross-subsidization. Deflate them with the cost-tocharge ratio. Productivity costs: costs associated with lost or impaired ability to work or engage in leisure activities due to illness or death. Also referred to as indirect costs - most are borne by patient or family. Discounting: a method that allows for differential timing of costs and consequences - the process of converting future dollars and future health outcomes to their net present value. A typical discount rate is 3%, with a range up to 7%. There are arguments about whether effects/benefits should also be discounted. Sensitivity Analysis: Calculations that isolate individual factors that are involved in an analysis to determine the degree to which they influence the outcome of the entire analysis. They specifically assess the uncertainty of the underlying assumptions by varying the key variables across a plausible range of values. To perform, the authors need to specify a plausible range of values for a variable - obtained from reviewing the literature, consulting an expert opinion, or using a specified confidence interval around the mean rate or cost. One-way analysis: vary one element, while holding all other variables constant. The most typical type of analysis. Nth-way analysis: vary more than one element at a time. As most elements are varied the results are harder to explain, so is usually limited to 2 variables. Threshold analysis: analyst varies a particular parameter over a range of values to determine if it would lead to major change in the conclusions. Used to explore what if scenarios when data are not directly available to address policy questions. Causal Inference: a cause is anything that produces an effect or a result. An understanding of causation justifies actions surrounding prevention, diagnosis, and treatment. We now address causation by using empirical evidence to come to a scientific judgment about the likelihood that a given exposure is a cause (probabilistic causation). Kochs Postulates: The organism must be isolated in every case of the disease (necessary); the organism must be grown in pure culture; the organism must always cause the disease when inoculated into an experimental animal (sufficient); the organism must then be recovered from the experimental animal and identified. Represent an example of deterministic causation - but is too restrictive for today. Necessary: The factor MUST be present for the disease to occur. Sufficient: Once the factor or factors are present the disease WILL occur. Examples: Rabies virus is both necessary and sufficient for human rabies; Mtb is necessary but not sufficient for clinical TB; smoking is neither necessary nor sufficient for lung cancer. Starting point in determining causation is the demonstration of an association between an exposure and disease. Statistical testing will assess the role that chance played in an association - the size of the p-value here has NO bearing whatsoever on whether the association represents a causal relationship. Hills Causal Criteria: 1. Strength of Association: a large effect (RR>5 or <0.2) is less likely to result from confounding or bias 2. Time Order: to be causal the exposure MUST precede the disease (key advantage to using cohorts over XS or CCS is that the exposure is known to occur before the disease). The only absolute criteria. 3. Dose-Response Effect: does risk increase with greater exposure?? 4. Specificity: is the effect of an exposure specific - does it only cause one effect? Most relevant to infectious disease applications, rarely applies to non-infectious diseases. 5. Biologic Plausibility: is a cause-effect relationship plausible on biologic grounds? Depends on the current biologic understanding of the condition. 6. Consistency: Finding the effect in more than one study, population, study design, or sub-group. Does the effect persist among different sub-groups within the study (internal consistency)? Have other investigators found similar findings (external consistency)? 7. Reversibility: Does changing the exposure result in a change in disease? 8. Analogy: Are there other established cause and effect relationships that have a similar exposure or disease (not a very strong criteria, but one that is commonly presented in the media).