Epidemiology & Evidence Based Medicine

What are the measures used to describe prognosis?

1) 5 year survival 2) Median Survival 3) Response to treatment (improvement/survival) 4) Time to recurrence 5) Case fatality 6) Disease-specific mortality

What impairs inferences regarding disease causation?

1) A long interval between causal factor and effect 2) The same effect can occur from other causes 3) More than one causal factor is required to produce the effect (However there have still been successes in proving a causal relationship. E.g smoking and lung cancer, HPV and cervical cancer, hantavirus infections and pulmonary syndrome)

What is the criteria for a potential confounder?

1) A variable must be associated with the outcome (a risk factor) independent of the exposure 2) A variable must be associated with the exposure but not a consequence of it (Example: Review Course (Exposure), USMLE Score (Outcome). There is a crude association between the two. Then a potential confounding variable could be grade point average

Identify the three ways that abnormal may be defined

1) Abnormal as unusual: Anything that is not average may be seen as abnormal (commonly observed vs unusual to see) 2)Abnormal as associated with disease: If there is a clinically meaningful change from good health (results may be normal until associated with disease or risk factor) 3) Abnormal as Treatable: If treatment is required or leads to better clinical outcome, results may be seen as abnormal

what kind of variables should be assessed as confounders?

1) Age: Some medical conditions may be seen more in certain age groups 2) Known Risk Factors: Whether people smoke/drink a lot can be related to a disease 3) Known Prognosis Factors: These are beyond risk factors.

What is the impact of bias on study results?

1) Bias can undermine ability to understand the association of interest 2) Bias introduces incorrect findings, minimizes or overemphasized the association (Bias must be considered even in the best of studies)

Case-Control Studies Weaknesses

1) Can't determine incidence directly; must estimate RR with odds ratio 2) Selecting Cases 3) Selecting controls 4) Uncertainty in exposure-disease time relationship 5) Bias potential

Number Needed to Treat (NNT)

1) Describes how many patients need to be treated to prevent one of them from experiencing the outcome 2) Inversely related to proportion of patients in the control group (i.e., untreated / baseline risk) who suffer an adverse event 3) Represents an easy number to understand, especially when comparing the effect of different interventions (1/Control Event Rate - Experimental Event Rate)

Relative Risk Reduction (RRR)

1) Describes the magnitude of the effect 2) Represents the percent reduction in the risk of the studied outcome achieved by use of the intervention 3) Often used because they tend to be impressively large; however may overestimate clinical relevance (Control Event Rate - Experimental Event Rate/ Control Event Rate)

Absolute Risk Reduction (ARR)

1) Describes the risk difference in outcome between patients who have undergone one therapy and those who have not (AKA: Attributable Risk or Risk Difference) 2) Tells how many patients are spared the adverse outcome as a result of having received the experimental rather than control therapy 3) Represents a more clinically appropriate way to look at difference (Control Event Rate - Experimental Event Rate)

Cohort Studies Strengths

1) Directly determine incidence and risk 2) Assess relationship between an exposure (i.e. risk factors) and many diseases/outcomes 3) Follow logic of clinical question, i.e., (if persons are exposed, do they get disease) 4) Measures exposure first, so reduces bias and good method when exposure is rare

Case-Control Studies Strengths

1) Excellent for rare or unusual diseases 2) Smaller in size, quick, easy and cost effective 3) Can evaluate multiple risk factors for one disease 4) Can use secondary data and test hypotheses

How to interpret CI in a study

1) Figure out what the null value or no effect means. In the case of a ratio, no effect usually means one (e.g. RR, OR). In the case of a measured variable, it usually means 0 (e.g. AR, average) 2) See if the value for no effect falls within or outside the 95% CI. If its outside then the result is statistically significant at the 0.05 level. If within the CI, then the result is not statistically significant

Causal Inference Take home messages

1) Its difficult to prove that an exposure causes disease 2) Randomized trials provide some of the best evidence for causation but are not always available 3) Guidelines have evolved to help in reasoning that an exposure causes disease, especially when information from a randomized trial is lacking

Diagnostic Hypothesis' and their descriptions

1) Leading hypothesis or working diagnosis (single best explanation of patient's illness 2) Active Alternatives (not as good as 1, but likely serious or treatable enough to be actively sought in this patient. 3) Other alternatives (not likely, serious or treatable enough to be pursued now, but not yet excluded) 4) Excluded hypothesis (Causes of the problem have been disproved)

Types of Trials

1) Randomized Trial : Cost a lot but it is very good quality with good results 2) Cohort: Not as expensive as RT, good quality, good results 3) Case control: Not as expensive as Cohort often good results (not the best quality) 4) Cross Sectional: Cheap with questionable results 5) Case Series:C heap poor results

Cohort Studies Weaknesses

1) Requires a large sample size (many more subjects are enrolled than experience of event interest) 2) Inefficient for rare diseases and takes a long time to complete 3) Expensive because of resources needed and validity affected by bias from subject attrition and loss to follow up (We may want rare exposure but we don't want a rare disease because the sample size would have to be massive just to find a few cases)

Risk Assessment

1) Target condition: Disease or health outcome that preventative care intervention avoids, identifies early or prevents from recurring (relative importance is assessed by its frequency and severity) 2) Risk Factors: characteristics that are either directly related or likely to lead to the target condition. May include demographic qualities, behaviors, environmental factors and system factors (relative importance is assessed by frequency and magnitude)

Criteria for building causal arguments

1)Exposure precedes development of diseases (cohort studies are great at showing this 2) Strength of association (high RR) 3) Dose-Response relationship is apparent (more exposure = increased risk of disease) 4) Consistency of findings from study to study

Criteria for Building Causal Arguments (Part 2)

5) Biologic plausibility (relationship makes biologic sense) 6) Other possible explanations have been considered (confounders) 7) Reduction or elimination of the exposure results in less disease 8) Findings are consistent with other studies or types of data including animal experiments

Say we work in the ICU and we want to know the incidence of nosocomial infections among our patients who were in the ICU for varying lengths of time in 2007. We found that 5031 patients remained under observation for a total of 127,859 patient-days. Since 596 patients developed an infection that met our case definition, the incidence rate can be estimated as

596/127,859 = 0.0047 cases/patient day or 4.7 cases per 1000 patient-days spent in ICU

Likelihood Ratio Enterpretation

>10 : Strong evidence to rule in disease 5-10: Moderate evidence to rule in disease 2-5: Weak evidence to rule in disease 1: No change in likelihood of disease 0.2-0.5: Weak evidence to rule out disease 0.1 - 0.2: Moderate evidence to rule out disease < 0.1: Strong evidence to rule out disease

Confounding Bias

A mixing of effects between the exposure, the disease and other factors associated with both the exposure and the disease such that the effects of the two processes are not separated. Occurs when two factors are associated with each other and the effect of a third factor confuses the association (Many risk and prognostic factors can be confounding variables if unequally distributed between groups. Confounding is a distortion of the exposure -disease association by the effect of some third factor)

Risk

A population concept A statement of probability that people who are exposed to certain factors will develop a particular disease more often than similar people who are not exposed The probability of a disease doesn't hold for an individual (who either gets the disease or doesnt) but probabilities in groups of people can guide clinical decision making in individuals

1) In a study of the relationship between smoking history and lung cancer, the investigators recruited all patients diagnosed with lung cancer at Hospital A between October 196 and October 1998. Controls were recruited from persons about the same age attending a health fair at the local mall during the same week in which cases were diagnosed. Cases and controls were interviewed regarding their past smoking history. a) selection Bias b) Measurement of information bias c) confounding d) chance.

A) Selection Bias (the way participants were being selected in the study

Dr. Sim works in a rural hospital emergency room and needs to quickly diagnose and stabilize a critically ill patient. she has only two relatively mediocre tests for immediate use. What will be the net effect of her diagnostic strategy if she uses both test together? a) Sensitivity will increase b) Negative predictive value will decrease c) sensitivity will decrease d) specificity will increase

ANSWER: A Sensitivity will increase because this is parallel testing and when you do this both sensitivity and negative predictive values increase

A cohort study is conducted to evaluate the relationship between dietary fat intake and development of breast cancer in women. In the study 200 women with high fat diets are compared with 200 women who are on a low fat diet. Both groups start at age 50-55 years and are followed for 15 years. During the follow-up period, 9 women in the high fat group and 6 women in the low fat group are diagnosed with breast cancer. What is the AR

AR = (9/200) - (6/200) = 0.015 or 1.5 per 100 women 95% CI (1.3 - 1.7) per 100 women This is there is excess risk of breast cancer attributable to high fat diets in the exposed women (95% CI does not include null value of zero, so result is statistically significant)

Interpreting Attributable Risk (AR)

AR = 0 No association between exposure and risk of disease ("no effect", "null value") AR > 0 Exposure is associated with excess amount of disease in the exposed AR < 0 Exposure is associated with decrease amount of disease in the exposed (We subtract to get the AR)

Cross Sectional Studies

Also known as prevalence studies are very commonly used methods. Exposure and disease status are assessed at the same time. Individual is unit of observation and analysis. Typically descriptive in nature to quantify magnitude of the problem (answers questions like is there a medical need in the community? Or should people who are not being treated be treated?) In this study design we can't say whether or not the disease or problem is the cause of something or the result of something

Implementation of Causal Criteria

Arriving at a tentative inference of causal or non causal is a subjective process Arguing a case for causation depends to a great degree on the strength of the research designs Associations are observed; Causes are inferred

What happens to the sensitivity and specificity if the test cut off point is shifted

As the cut off point increases the true positives and false positive will be smaller and the false negatives and true negatives will get bigger. So shifting the cut off point affects the ability for you to find something. Low cut off points are used in populations where diseases are more prevalent to pick up all of the cases

Sources of Variation

Biological Variation 1) Within Patients (if you measure BP at different times in the day you may get various readings. 2) Between Individuals (some people have naturally high BP while others have naturally low) Measurement Variation 1) Instrument (you can get inaccurate measurements if the instrument is not calibrated or if the BP cuff it too big for example) 2) Observer (if you have someone thats really good at taking BP vs someone who is bad then measurements may vary) (There will always be variation when taking measurements. The more variations you have the more likely you are to have bias in your study)

Blinding/Masking in Randomized Trials

Blinding: Investigators and/or study participants assess or measure the outcome without knowing which intervention a patient is receiving Levels of blinding include: Those who allocate patients to treatment groups, patients in the study, physicians and investigators and researchers who assess outcomes (Blinding adds another level of quality and validity to the study because the patients don't know who is getting what)

Risk vs Prognosis

Both involve a set of probabilities of various outcomes over time. A distinction should be made between factors associated with increased risk of developing a disease (risk factors) and those that predict a worse outcome once the disease is present (prognostic factors)

What is one way we can tell normal from abnormal in patients results?

By using pictures of continuous variables that show the center of the data, the variability and the shape of the distribution. If we can assume the data follows a normal distribution, then we can get an idea of how the data is spread out for a continuous clinical variable

Case Reports/Case Series

Case Reports: Describe the experience of a single patient with an interesting finding (e.g. a man with febrille illness that developed from a tick bite) Case Series: Describes the experience of a group of patients with similar diagnosis. (e.g. over a period of 18 months 65 people were seen with the same rare symptoms) There is no comparison group only the case(s) being described. Therefore, it is susceptible to bias and it can't be used to make treatment decisions)

Diarrhetic shelfish poisoning (DSP) is an acute gastrointestinal illness. In July 2011, a state health department received a report of a family who experienced illness after consuming recreationally harvested mussels. Three family members experienced symptoms beginning 4, 7, and 15 hours after consumption of cooked mussels, respectively. A fourth adult family member who consumed mussels did not become ill. Signs and symptoms included vomiting, diarrhea, body aches, fever and chills; no neurologic symptoms were described, This report resulted in closure of harvest areas and a recall of commercial shelfish products. This type of study is a : a) Case Series b) Cross-Sectional c) Case-Control d) Cohort e) Randomized Trial

Case Series (No comparison group, we are talking about a group of people who got sick)

In 2004, women with recently diagnosed ovarian cancer (n=168) and comparison group of women (n=251) were interviewed about symptoms over the previous 6 months. Those with cancer were interviewed on average 4 to 5 months after diagnosis. The authors found significant differences in symptoms between ovarian cancer patients and comparison group with bloating, lack of appetite, abdominal pain, fatigue, urinary frequency and constipation occurring significantly more frequently in ovarian cancer cases than in the other group of women. The type of study design best described by the scenario is? a) Case Series b) Cross-Sectional c) Case-Control d) Cohort e) Randomized Trial

Case-Control

where do questions come from in EBM?

Clinical findings, etiology, differential diagnosis, diagnostic tests, prognosis, therapy, prevention and self-improvement

Notes on NNT

Clinicians mis-estimate baseline risk and effects of therapy. NNT is easily calculated from the risk difference (ARR) Awareness of threshold NNT can help anticipate the risk reduction to look for in a therapy Patients want to know their personal absolute risk reduction, so clinicians need the NNT for a group of individuals resembling their patients

To better understand the association of childhood risk factors for heart disease on adult mortality, a group of 4867 American Indian children without diabetes, from reservations in Arizona born between 1945-1984 were assessed for body mass index, glucose tolerance, blood pressure and cholesterol levels. Children were grouped into high to low risk quartiles for each risk factor and followed for a median of 23.9 years to measure premature death (defined as death before age 55 years of age). The type of study design best described by this scenario is: a) Case Series b) Cross-Sectional c) Case-Control d) Cohort e) Randomized Trial

Cohort

Other Screening Biases

Compliance Bias: Compliant patients tend to have better prognosis regardless of preventative activities Volunteer Bias: People who show up to volunteer are different from those who don't Selection Bias: Occur if the subjects are permitted to choose whether to go into a drug group or placebo group rather than being assigned randomly Over-diagnosis Bias: Occasionally you will have someone over zealous and determined to find disease and they may start to over diagnose

A study compared two groups on outcomes for a chronic clinical condition. The distribution was: Treatment group: 70% are non smokers; 30% are heavy smokers; control group: 35% are non smokers; 65% are heavy smokers. Treated patients appear to have superior outcomes to untreated patients. Which is the best explanation? a) Selection bias b) Information measurement Bias c) Confounding d) chance

Confounding (imbalance between the amount of smokers in each group)

Residents of three villages with three different types of water supply were asked to participate in a survey to identify cholera carriers. Because several cholera deaths had occurred in the recent past, virtually everyone present at the time submitted to examination. The proportion of residents in each village who were carriers was computed and compared. The type of study design best described by this scenario is? a) Case Series b) Cross-Sectional c) Case-Control d) Cohort e) Randomized Trial

Cross-Sectional (because its a prevalence study, 3 villages, 3 water supplies but we are trying to get an idea of what is going on to see if there is any kind of association between cholera and water.

Notes about different rates (standardized rates/adjusted rates)

Crude rates are not useful for comparing groups It may be necessary to extract out demographics (age group, sex, preoperative risk) first in order to compare the rates Use a standard population to control for the effects of demographics and allow for a valid comparison of rates (as if the two groups had the same demopgraphics

Survival Curve Calculations

Curve starts at 1.0 on the vertical line, continues across until death occurs. Dropouts with incomplete information have no immediate effect on height of curve. A death at time t causes curve to drop to the following amount: (Current height) x (N-1)/N (N=number of survivors still under observation just prior to time t)

Incidence

Describes the proportion of a population, initially free of the outcome of interest, that develops the condition over a given period of time Refers to new cases of disease or new outcomes.

What are the issues in Randomized Trials?

Did patients get the treatment they were supposed to?(Adherance, crossovers, losses) Did patients get additional treatments? (Cointerventions - prescription/over the counter) Were there other differences between patients and controls? (Compliance) Were outcomes assessed without bias? (Blinding) Are the outcomes appropriate, without short term substitutes? Are the primary outcomes presented, not only the subgroup outcomes?

Principles Underlying Causal Inference

Distinguish between association vs causation. Criteria for formulating a plausible causal hypothesis based on results of either randomized studies or non randomized studies. (Disease it rarely due to a single cause)

Lead Time Bias

Early detection of disease is confused with increased survival If you are screening you may think that because you caught the disease early they will have increased survival. However, this may not be true and they may live just as long as people who didn't have the screening. (the way to look at this is to look at survival age at death)

A double blind randomized trial assigned working adults at a US manufacturing company to receive either trivalent inactivated influenza vaccine or sterile saline injection. Rates of influenza like illness, lost workdays and physician visits during the following year were compared. This study would most likely be considered more of a? a) efficacy b) effectiveness c) efficient d) explanatory

Effectiveness (they were looking at a large company work site so this is more like a real life situation. Not an ideal situation. Here they want to see if this vaccine would work in a general population.) Efficacy would be a situation where they would want to try out a new vaccine, the study would be a small tightly controlled clinical trial at the pharmaceutical company with about 10-20 people. Some people would get the vaccine and some would get the placebo and the patients would be under close supervision to see if Igs form. Then if it works it would be released out to the general population. So both efficacy and effectiveness are both necessary.

Efficacy vs Effectiveness (Judging Treatment Results)

Efficacy: Does the treatment work in an ideal setting? Treatment has desired effects in those who receive it under ideal circumstances. Includes compliant patients so analysis can be either intention to treat or explanatory trial Effectiveness: Does the treatment work in ordinary circumstances; usual patient care? The results are what would be experienced by most patients in regular care. Usually analyzed by intention to treat

Define Epidemiology and Clinical epidemiology

Epidemiology: The study of the distribution and determinants of health related states or events in specified populations and the application of this study to the control of health problems Clinical Epidemiology: The science of making predictions about individual patients by counting clinical events in groups of similar patients and using strong scientific methods to ensure that the predictions are accurate

Confidence Intervals

Expresses statistical precision around the point estimate by showing the range of values likely to include the true effect size; it describes how confident we are that the true point estimate falls in the range Interpretation of 95% CI: If the study is unbiased, there is a 95% chance that the interval includes the true point estimate 95% CI = Point Estimate +/- 1.96 (SEM)

When looking at patients and their symptoms it is important to look at the probability that a patient may have a particular disease based on their population's characteristics

For example a 35 year old female that presents with chest pain. Normal cholesterol, non smoker and no family history of disease will probably only have a 10% probability of having CHD vs a 65 year old male with chest pain. He smokes, he has type II diabetes and very high cholesterol. His probability of having CHD is about 90%

When doing a risk assessment, the relative importance of a risk factor is assessed by: 1) Frequency and Severity 2) Frequency and Magnitude 3) Severity and Probability 4) Severity and Magnitude

Frequency and Magnitude (when we are looking at the risk assessment and the target condition we look at frequency and severity)

Predictive Value

Given a positive or negative test, how likely is that to be a true result or Of all the people with a positive or negative test, what percentage actually have or don't have the disease?

Explain the test characteristics of validity and generalizability in providing patient care

How does the test perform compared to the gold standard? Validity (Accuracy): Did the test measure what it was supposed to measure and decrease bias (Internal) Generalizability: The degree to which the results of a study based on a sample can represent the results that would be obtained from the entire population where the sample was drawn

Likelihood Ratios (LR)

How much more likely is the test to be positive among those with disease as apposed to those without disease. Uses Odds (probability of the event/ 1-probability of the event) (Can be used for multiple levels of the cut point and then compared)

Power and Sample Size

If a study has low power and finds no effect, you don't know if no effect is present or if the study lacked power to confirm the effect (ideal power would be 0.8) If power is held constant , the greater the treatment effect, the fewer the patients are needed New treatments that only improve over traditional treatments by 25% or so can require large sample sizes to demonstrate significant improvement (avoid a power problem by doing a sample size calculation ahead of time)

Continuous Test Results and Cut Off Point

If a test result is continuous, a cut off point must be selected for disease status. A cut off point is usually set by the test manufacturer.

Notes on sensitivity and specificity

If the test is good enough, use the test characteristics of sensitivity and specificity to help diagnose patients once you get the results back When you dont have a test with good specificity or sensitivity you use likelihood ratios

Notes on Risk Reduction (Balancing Benefits and Risks)

If you can estimate a patient's risk of outcomes without treatment, then you can tailor a clinical trial data about treatment results to an individual by calculating risk differences Presenting data as risk differences makes the benefits and harms of intervention easier to compare.

A study is conducted to evaluate the relationship between dietary fat intake and the development of prostate cancer in men. Men are at least 65 years of age at enrollment; they were asked about usual diet by questionnaire. Of 100 men reporting a high fat diet, 7 of them developed prostate cancer during the study follow up period. The risk of prostate cancer among men in the high fat diet intake group is a measure of: a) Prevalence b) Specificity c) Incidence d) Variation

Incidence

Cohort Study Incidence Density and Person-Time

Incidence Density E+ = a/PTE+ Incidence Density E- = c/PTE- Relative Risk (Rate Ratio) = [a/PTE+]/[c/PTE-]

Cohort Studies: Measures of Association

Incidence Risk: Number of newly identified cases during follow up (l1=incidence in exposed, lo=incidence in non-exposed, lt=incidence total) Risk Ratio (Relative Risk) RR: l1/lo (if a positive association exists between exposure and disease, we would expect the incidence of disease in exposed to be greater than incidence of disease in non exposed) Attributable Risk (AR) = l1 -lo (if positive association exists between exposure and disease we would expect greater incidence in exposed persons

The Two Approaches to Analysis of Results

Intention To Treat: Analyze according to group assigned by randomization regardless of whether they actually received treatment (best option because it preserves randomization) Explanatory: Analyze according to treatment actually received, regardless of randomization (For both of these we are trying to figure out if treatment made a difference)

Cohort Studies (Longitudinal Studies)

Key Comparison: Risk Factor vs No Risk Factor. Group of people who have something in common; observed over time to see what happens to them. Starts with people free of outcome of interest but with varying risk. Moves from potential cause to effect Useful when the exposure may be or is known to be harmful and could not be randomly assigned by an investigator. (if previous information is available on exposure, can use historical data to go forward in time)

LR+ vs LR-

LR+: Probability of a positive test result for a person with disease/ probability of a positive test result for a person without disease. Use when test is positive LR+ = sensitivity / 1-specificity LR-: Probability of a negative test result from a person with disease / probability of a negative test result for a person without disease. Use when test is negative LR- = 1-sensitivity / specificity (if a test has both high specificity and sensitivity the LR+ will be a big number)

Measurement Bias (Information Bias)

Measurement of information is consistently dissimilar in different groups of patients or data are collected differently for subjects in different groups in a way that could affect the outcome of the study

A study classified persons as exposed or unexposed on the basis of alcohol exposure. However, errors were made in measuring the amount of alcohol consumed. Some subjects with high alcohol intake were classified as low non drinkers (e.g subjects who wish to impress the interviewer and report the amount of alcohol intake they feel is socially acceptable). Misclassification of exposure led to the finding of a very small increase in risk of cancer, with non significant confidence intervals. a) selection bias b)measurement/information bias c) confounding d) chance

Measurement/information bias

Incidence Density (person-time)

Measures the number of new cases in a dynamic population with people entering and leaving Provides the density of new cases of disease in time and place using person-time at risk for the outcome event Probability of an individual developing the disease during a specific period of time using total person time as the denominator I density = number of new cases of disease during a given period of time/ total person time

Cumulative Incidence

Measures the rate of new events in a group of people of fixed size. New cases of disease are accumulating over time. Denominator is all susceptible people without the disease at the beginning of the time period aka the probability of an individual developing the disease during a specific period of time using number of persons at risk for the denominatior

Survival Curve

More informative than summary numbers since it portrays timing of events. Used to determine which treatment works better over said period of time. Area under the curve is the number of person-years lived by the population studied Probability of events is dependent on the length of follow up The median survival and 5 year survival percentages can be found using the curve

Infant Mortality Rate

Number of deaths < 1 year of age (in 1 year) / Number of live births in that same time In a certain country with a population of 6 million people, there were 42,000 live births during the year ending December 31, 2002. During that time period, there were 4,000 deaths in infants < 1 year of age. Infant Mortality Rate = 4,000/42,000 = 0.952 or 9.5 per 100

Proportionate Mortality Rate

Number of deaths from a specific disease / total number of deaths during the same time period In a certain country with a population of 6 million people, 85,000 deaths occurred during the year ending December 31, 2002. 3,000 deaths were due to malaria 5,500 deaths were due to respiratory infections 40,000 deaths were due to heart disease 36,500 deaths were due to other causes Proportionate mortality of deaths from malaria =3,000/85,000 = 0.035 or 3.5%

Age Specific Mortality Rate

Number of deaths in specified age group in time t / Population in that age group at that time A certain country has a population of 6 million people, comprising 1 million under age 5 years and 1.5 million age 5-10 years. During the year ending December 31, 2002 there were 7,000 deaths that occurred among children under the age 5 years while 4,300 deaths occurred among children age 5-10 years. The age-specific mortality among children < 5 yrs = 7,000/1,000,000 = 0.007 or 7 per 1000

Disease (Cause) Specific Mortality Rate

Number of deaths of specified cause in time t / Total population at the same time period (at risk of death) In a certain country with a population of 6 million people, 85,000 deaths occurred during the year ending December 31, 2002. 3,000 deaths were due to malaria The malaria-specific mortality rate = 3,000/6,000,000 = 0.0005 or 5 per 10,000

Crude Mortality Rate

Number of new deaths in time t (usually 1 year) / Total population during that time In a certain country with a population of 6 million people, 85,000 deaths occurred during the year ending December 31, 2002. The crude mortality rate = 85,000/6,000,000 = 0.0142 or 14.2 per 1000

Case Fatality Rate

Number of people who die from disease / Number of people who have the disease In a study of 200 patients with MI who were treated at a university hospital and then followed to determine their clinical status one year after diagnosis, 120 patients had unresolved deficits and 40 patients had deaths attributed to the MI. The 1-year case fatality = 40/200 = 0.20 or 20%

Interpreting Odds Ratios (OR)

OR = 1 No association between exposure and likelihood of disease ("null value", "no effect") OR > 1 Odds of exposure in cases is greater than odds of exposure in controls OR < 1 Odds of exposure in cases is less than odds of exposure in controls

Study Designs

Observational: Can be descriptive and include case report, case series or cross sectional studies or it can be analytical and include a cohort or case control Experimental: Includes randomized trials which include clinical trials and field trials

Case-Control and The Odds Ratio

Odds of exposure if case = [a / (a+c)] / [c / (a+c)] = a/c Odds of exposure if control = [b / (b+d)] / [d /(b+d)] = b/d Odds of exposure given disease = (a/c)/(b/d) = (ad)/(bc)

Tertiary Prevention

Once a disease has developed and has been treated in its acute clinical phase, tertiary prevention seeks to soften the impact caused by the disease on the patient's function, longevity, and quality of life. Example: Cardiac Rehabilitation following a MI, follow up exams to observe re-occurrence of cancer in cancer patients

Parallel vs Serial Testing

Parallel Testing: This occurs when we order several test at once. It is used for rapid assessment situations, It maximizes (increases) sensitivity and negative predictive value Serial Testing: Order next test on the basis of prior test results. Useful in clinical situations when assessment can be done over time or tests are expensive. Maximizes specificity and positive predictor value

Characteristics of diagnostic and screening situations when choosing tests

Patients are very different when looking at diagnosis vs screening. The patients you are screening will have little to no symptoms vs the patients you are diagnosing Prevalence will be higher in the patients you are diagnosing vs screening

Comparison groups in randomized study

Placebo controls if no accepted treatment exists, Usual care or intervention as control or another intervention Assignment of patients into groups without bias and a method of controlling for confounders. Random assignment includes simple, block assignment and randomization of pairs of subjects Issues include: Unpredictable outcomes, placebo effect or regression to the mean

Notes on Point Estimates and Confidence Intervals

Point estimates and confidence intervals can be used to describe the statistical precision of any rate, comparison of rates or other summary statistics Confidence Intervals provide more information than p values alone because they show a range of possible true values CI gives a perspective concerning sample size and power and includes information about statistical significance

Positive Predictive Value (PPV) and Negative Predictive Value (NPV)

Positive Predictive Value: Used when the test is positive and it is the percentage of those with a positive test who have the disease (TP) PPV = TP / (TP + FP) Negative Predictive Value (NPV): Used when the test is negative and it is the percentage with a negative test who do not have the disease (TN) NPV = TN / (FN + TN) (Prevalence and PPV are directly proportional. As Prevalence increases PPV increases and vice versa)

Pre-test and Post-Test Likelihoods

Pre-test Likelihood = Prevalence Post-test Likelihood = Positive Predicitve Value Post test likelihood give a positive test: Same as PPV (Probability that patient has disease given a positive test) Post test likelihood given a negative test: 1 - Negative Predictive value (probability that patient has disease despite negative findings

The entire term 1 class in SOM was required to be vaccinated for Hep B upon entry. At orientation they were asked about other vaccinations. Out of 540 students, 218 reported being vaccinated for influenza. This is a measure of: a) Prevalence b) Incidence c) Risk Factor d) Positive Predictor Value

Prevalence

A new test has been developed to detect a hypothetical chronic disease. A total of 500 patients were referred to a laboratory for testing. There were 200 diagnosed cases (by gold std) of disease among the 500. The test yielded 100 positives, of which one-fourth were true positives. Calculate the following: Prevalence, Sensitivity, Specificity, PPV and NPV

Prevalence = 200/500 = 40% Sensitivity = 25/200 = 12.5% Specificity = 225/300 = 75% PPV = 25/100 = 25% NPV = 225/400 = 56%

Disease Frequency

Prevalence = Incidence x Duration P = I x D (Remember that risk goes with incidence and not prevalence. So though the prevalence of TB may be higher in a population, if another population has a higher incidence then the risk of TB is higher in the later population)

Prevalence (Pre-Test Likelihood/Probability) (On the basis of population characteristics)

Prevalence: The percentage of people at one given time that have the disease Prevalence = (TP + FN) / (TP + FN + TN + FP) (It is important to know the prevalence of a disease before you order a test to see how big the probability of your patient having the disease is)

Secondary Prevention

Procedures that detect and treat pre-clinical pathological changes and thereby control disease progression. Example: Screening

Interpreting Relative Risk (RR)

RR = 1 No association between exposure and risk of disease ("null value", "no effect") RR > 1 Exposure is associated with increase in risk of disease RR < 1 Exposure is associated with decrease in risk of disease (If the confidence interval includes the null value of 1.0, then the finding could be due to chance, and we say the result is not statistically significant. p-value = >.05)

Summary on Event Rates

RRR is constant across populations ARR varies with underlying risk The lower the event rate in the untreated (CER) the larger the difference between RRR and ARR If ARR is large, we need to treat few patients to observe benefits in some

Difference between RRR and ARR in the interpretation of treatment

RRR is often more impressive than absolute risk however ARR is better to use when explaining to patients The lower the event rate in the control group, the larger the difference between RRR and ARR (There will be a bigger separation between low risk people RRR and ARR vs high risk people RRR vs ARR)

How can you control confounding in the design stage?

Randomization: Ensure known and unknown confounders are evenly distributed in study groups Restriction: Limit subjects to one category of a confounder e.g. only men; no variability, so no confounding Matching: Enroll controls to be similar to cases on the potential confounders; however it must follow matched analysis

Randomized Trial

Randomized trials allow us to determine if a treatment works and how it changes the course of the disease Sampling includes inclusion and exclusion criteria, patients in clinical trials are usually a highly selected biased sample of all patients with the condition of interest. Intervention characteristics include generalizability (can it be done in more than one setting), complexity (is it too complex for people to want to be apart) and strength (findings should be strong enough to say the study was worthwhile)

Ratio vs Proportion vs Rate (Frequency Measures)

Ratio: Value obtained by dividing one quantity by another (e.g. ratio of male to female births in U.S in 1979: 1,791,000/1,703,000 =1.05) Proportion: A ratio where the numerator is always part of the denominatior (e.g. proportion of males among all births in 1979: 1,791,000/3,494,000 = 51.3%) Rate: A change in one quantity per unit change in another over time. (e.g. rate of developing STI during a 1 year study among 250 persons. 6 new cases of STI diagnosed/250 persons = .024 or 2.4 per 1000 persons per year)

Primary Prevention

Seeks to prevent the onset of specific diseases via risk reduction: by altering behaviours or exposures that can lead to disease, or by enhancing resistance to the effects of exposure to a disease agent. Examples include smoking cessation and vaccination.

Case-Control Methods

Selection of Cases: Case definition is very important, incident (new) cases are best. All cases should have an equal probability for selection as a case Selection of Controls: Identical in every respect except disease of interest; cannot be chose in relation to the exposure of interest. Can be population, community or hospital based and can have multiple controls per case. (Case-control Studies can be done within a cohort study)

Define sensitivity and specificity

Sensitivity: Proportion of individuals with the disease who have a positive test (True Positives) Sensitivity = TP / (TP + FN) Specificity: The percentage of individuals without the disease who have a negative test (True Negatives) Specificity = TN / (FP + TN) (In all settings and patients, the sensitivity and specificity of a test tell us about the validity of the test and they don't change (inherent characteristics of the test)

Length Time Bias

Slow developing conditions are more likely to be picked up in screening For example time development for tumor. Some tumors develop slowly than other so the faster developing conditions would be harder to detect in screening

SNOUT and SPIN Rules

Sn (ensitivity) Nout (SNOUT) : If sensitivity is high, then you can RULE OUT the disease if the test comes back negative Sp(ecificity) Pin (SPIN): If specificity is high, then you can RULE IN disease if the test comes back positive

Assessment of Outcomes in Experimental Design

Specify outcomes before conducting the study Compare outcomes using RR, AR, OR Summarize treatment effects by determining change in the outcome resulting from the intervention: Relative Risk Reduction (RRR), Absolute Risk Reduction (ARR) and Number Needed to Treat (NNT)

Statistical vs Clinical Significance

Statistical Significance: Is this difference likely to be real (beyond random variation) p-value <0.05 Clinical significance: IS this difference likely to be clinically important? Both are influenced by sample size. If small, clinically significant differences may not reach statistical significance. If large, clinically insignificant results might be found to be statistically significant (Clinical significance is more of a judgment call. For example, if blood pressure changes by 2 mmHg that might be statistically significant in a big study however that is not necessarily clinically important)

Controlling confounding in data analysis (Adjusting for confounding-if you don't find it early on)

Stratification: Separating data into strata of confounder (people who are extremely healthy to people who are average) Mantel Adjusted RR: Weighted Strata Multivariate Analysis/Modeling: Advanced statistical techniques, logistic regression (don't focus to much on detail here just know that there are ways to check for confounding and adjust for it in the study)

Strengths and Weaknesses of Randomized Trials

Strengths: Control over study situation, control of exposure (timing and frequency) and random assignment. Weaknesses: External validity-will it work in the real world?, Adherence to protocols and ethics of withholding treatment

what are the strengths and weaknesses of a cross sectional study?

Strengths: Program planning and justification, may generate new etiologic hypothesis, relatively easy and inexpensive, can use data collected for other purposes Weaknesses: No cause-effect, measures at one point in time; no temporality and prevalent cases are survivors These are good comparison studies

Strengths and Limitations of Survival Curves

Strengths: Provides information about patients entering studies at varying points in time; increases efficiency of studies Limitations: Potential for selection bias if non random selection to receive a treatment or loss to follow up (Frequently used to present data addressing the time until an event occurs (e.g death, recurrent disease, fracture, MI)

Prognosis

Studies of outcomes in patient populations (outcomes can either be poor or good)

Selection Bias

Systematic error in a study from procedures used to select subjects or error from factors that influence study participation and follow up of persons in the study

Bias

Systematic errors in collecting or interpreting data such that there is deviation of results or inferences from the truth Bias results from systematic flaws in study design, data collection, or the analysis or interpretation of results

It is know that the prevalence of Condition C is about 20% in the community. Assume you have two tests for diagnosing condition C in persons who present with headaches and other symptoms. Test A has a LR of 5 and Test B has a LR of 10. If used in this setting and your symptomatic patient has a positive result for Test A and Test B. what is the post test probability of condition C for each test? (Use nomegram)

Test A is 54% and Test B is 70%

Test characteristics

Test may be pathology, clinical findings, symptoms or a combination The characteristics of a diagnostic test describe its validity in several ways Most frequently: Sensitivity & Specificity

Notes on PPV

The PPV of a test is closely related to the prevalence of the disease in that population; and to a lesser degree the sensitivity and specificity. So the higher someone's risk of having a disease, the higher the PPV of the test and the lower the NPV of the test

Point Estimates

The effect size observed in a particular study. It is the best estimate from the data of the true (but unknown) effect size in the population (For example the odds ratio in a case-control study, the relative risk in a cohort study or the relative risk ratio in a randomized control study)

What is the relationship between the number of tests ordered and the percentage of normal people with abnormal results?

The more test you order the higher the probability of having abnormal results in normal people (For example 1 ordered test may give you 5% normal people with at least one abnormal results vs ordering 20 test and having 64% of normal people with at least one abnormal result. Basically, if a physician orders enough test false positives will be discovered in virtually all healthy patients)

Differential Diagnosis

The process of weighing the probability of one disease vs that of the other diseases possibly accounting for patient's illness. (Test and follow ups usually help when eliminating differential diagnosis)

The relationship between incidence and prevalence

The relationship between incidence and prevalence depends greatly on the natural history of the disease state being reported. In the case of an influenza epidemic, the incidence may be high but not contribute to much growth of prevalence because of the high, spontaneous rate of disease resolution. In the case of a disease that has a low (or zero) cure rate, but where maintenance treatment permits sustained survival, then incidence contributes to continuous growth of prevalence. In such cases, the limitation on prevalence growth is the mortality which occurs in the population. (Prevalence will continue to grow until mortality equals or exceeds the incidence rate)

How do you study risk?

There are rigorous methods to obtain estimates of risk 1) Observe relationship between exposure to a possible risk factor and subsequent development of disease 2) Count clinical events in groups of similar patients 3) Compare the risk of disease in persons with certain factors and without the factors 4) Look at absolute (attributable) and relative measures of risk

Receiver Operating Characteristics Curve (ROC)

These are ways of estimating the best cut off point for a test measured on a continuous scale. They are also a way to compare the performance of two test. the test with the highest area under the curve is the best one to use (best curve hugs the upper left corner and give us maximum specificity and sensitivity) Sensitivity is on the y axis and specificity is on the x axis

Case-Control Studies

They all have a key comparison (disease vs no disease). They find all cases that meet a defined criteria and choose a representative group of controls that are just like the cases except they dont have disease They measure the risk of exposure in the disease group vs the no disease group

The impact of regression to the mean on testing situations

This is the natural tendency for a variable to change with time and return towards population average. As a result, clinically it may be best to use the average of the repeated measures rather than a single measure. It is also important not to rush to treat abnormal results as a condition and just observe it over time. (For example: when measurement of BP levels in a group of people are repeated, many values converge (regress) towards the mean)

How can a systematic error be introduced into a study and cause bias?

Through selection bias Through measurement/information bias Through confounding bias The study may also be incorrect because of random variation or chance

An intention to treat analysis is according to the treatment a) that patients actually receive b) that patients hoped to get c) to which patients were initially assigned d) to which patients personal physician intended them to get

To which the patients were initially assigned

Chance

Unlike bias, chance is a random variation. Results are likely to be abnormally high as low.

How do you assess the effects of an intervention to decide if the treatment will be useful for your patient?

Use Absolute Risk (EER and CER) to get 1) Relative Risk Reduction (RRR) 2) Absolute Risk Reduction (ARR) 3) Number Needed to Treat (NNT) (EER = Experimental Event Rate and CER = Control Event Rate)

Gold Standard (Reference Standard)

Usually an invasive and/or expensive test that provides a high degree of accuracy. The best definition of whether the disease is truly present. (Most times we are not able to use the gold standard and have to resort to other test)

Define Validity and Reliability in terms of regarding measurement

Validity (Accuracy): Extent that the measurement represents what it is supposed to. This can be compromised by systematic errors or bias Reliability (Consistency; Precision): Extent to which repeated measurements are similar. This can be compromised by random error/chance

Measuring Exposure in Case-Control Studies

Validity of C-C studies also depends on avoiding bias in measurement of exposure/risk factors Presence of the outcome may affect the subject's recollection of the exposure (a mother with a child that has cancer may be more thorough in her thoughts and responses than a control mother with a child that doesnt have cancer) Ensure all subjects are asked the same questions, same level of detail, probing etc