Pop med

Prevalence of disease in population

probability that a test will be positive when applied to an individual who actually HAS the disease    a/(a+c)

• probability that a test will be negative when applied to a disease free individual
• d/(b+d)

c/(a+c)

b/(b+d)

c/(c+d)

b/(a+b)

• probability that an individual with a positive test actually has the disease.  
• DEPENDS ON PREVALENCE
• a/(a+b)

• overall frequency of correct diagnosis
• true positive + true negatives / total results

• "believe the positive" increases sensitivity
• "believe the negative" increases specificity
• first test becomes prevalence for second test in series.

study result that differentiates a positive from a negative finding.  Changing it changes the specificity and sensitivity.

statement of belief about population parameters

hypothesis the researchers wish to evaluate

belief that the treatment didn't do anything (opposite of alternative hypothesis).

a standard for rejecting the null hypothesis (P<0.05)

information evaluated to reach a conclusion

• truly innocent but judged guilty (false positive?)
• probability of detecting a significant difference when the treatments are actually equally effective.
• as probability of type I errors increase, probability of type II errors decreases

• truly guilty but judged innocent (false negative?)
• as probability of type I errors increase, probability of type II errors decreases

• (exposed sick/not sick) / (not exposed sick/not sick)
• [a/b] / [c/d] = ad/bc
• same as risk ratio in RARE diseases

• probability of detecting a predefined clinically significant difference in a study
• 1-beta
• type II error

• type 1 error, alpha
• probability of detecting a significant difference when the treatments are actually equally effective.
• risk of false positive

• division of 2 numbers.  numerator is ALWAYS in the denominator
• quantities of same nature

• division of two numbers
• numerator is NOT in the denominator
• compare quantities of different nature

• division of two numbers, time in the denominator
• Speed of occurrence of an event over time

• proportion of the population that acquire or develop a disease in a period of time
• probability of developing a disease (over a period)
• NEW CASES/population at the beginning

• proportion of the population that acquire or develop a disease in a period of time
• speed of developing a disease
• NEW cases/total animal-time of observation

• cumulative incidence during an outbreak, for ENTIRE epidemic period
• not really a rate but a proportion
• new cases/population at the beginning

• proportion of a defined group or population that has a clinical condition or outcome at a given point
• cases/population     at a POINT in time
• focus on EXISTING states, not new
• aka point prevalence, prevalence proportion, prevalence rate

• subset of the population in given disease state
• removed by death or getting better, or leaving risk population

• combination fo prevalence and incidence
• cases of disease existing at beginning of study + new cases over study time / entire population

• probability that an event will happen/probability that an event will not happen
• case/non-case  (probability / probability)

• [a/(a+b)] - [c/(c+d)] 
• = grp 1 cumulative incidence - grp 2 cumulative incidence

[a/(a+b)] / [c/(c+d)] = a(c+d) / c(a+b)

no comparison group in descriptive

• aka hypothesis generating studies
• case report: description of rare dz in individual/population
• case series: describes a series of similar cases
• survey/census: describes a characteristic of pop.  survey is from a sample of pop, census is from ALL members of pop
• advantages: hypothesis generation, info about rare dz, characterizes disorders
• disadvantages: can't study cause/effect, can't assess disease frequency (except survey/census)

• hypothesis testing or explanatory studies
• observational: no individual intervention.  Tx or exposure in "non-controlled" environment.  Can observe concurrently, prospectively or retrospectively.  Also can be CROSS-SECTIONAL (single point in time for common dzs), CASE-CONTROL STUDIES (compare known disease to find exposure differences), COHORT STUDIES (compare known risk factor individuals to those without, evaluate incidence)
• experimental: can't control exposure, random assignment to groups, clinical trials most well known, ultimate step in causal hypothesis
• eval of diagnostic tests, reviews (systematic or meta-analysis).
• to start: question, comparisons (exposure/outcome), sample size, selection criteria, bias

• type of observational study (analytical)
• determines disease and exposure at a single point in time
• advantages: good for common, long-duration conditions, low cost, short time-frame for study, evals multiple exposures/outcomes, can calculate odds ratio
• disadvantages: prevalence, not incidence, not good for rare or newly emerging, don't know when dz occurred often, temporal sequence difficult to determine.
• good for cats/kidney disease

• type of observational (analytical) study
• compares individual with known disease status to find differences in exposure (ID cases first, then ID control group).  Data collected retrospectively
• advantages: good for rare dz, low expense, short time for study, ID multiple risk factors, smaller sample sizes
• disadvantages: increased bias (cases may not be representative, hard to find appropriate control), not good for rare exposures, only study one outcome, only calculate odds ratio, not risk ratio.  
• dogs with hip dysplasia

• type of observational (analytical) study
• compares individuals with known risk factor or exposure to individuals without risk factor or exposure (select subjects before disease, exposure determined before disease).
• evaluate risk over time (incidence), data collected prospectively.  
• advantages: direct estimate of effect (incidence, relative risk), temporal sequence established, decrease bias, eval multiple outcomes, GOOD FOR RARE EXPOSURES, risk ratio and odds ratio calculated
• disadvantages: expensive, no good for rare DISEASES or long latency, lots of subjects needed, can take a long time to complete, loss to follow-up, change in exposure over time

• type of experimental (analytical) study
• subjects randomly assigned to groups to control bias (uniform groups, environment, exposure).  
• Goal: only difference is experimental treatment.  
• PROSPECTIVE, RANDOMIZED TO GROUP, FOLLOW OVER TIME
• advantages: "gold standard, straightforward evaluations, most convincing evidence of correlation.  
• disadvantages: expense, inappropriate for some types of questions

be familiar

The quality of being logically or factually sound

• Prejudice in favor or against one thing, person, group compared with another
• systematic error in data
• makes a factor seem important when it isn't or opposite.
• Selection bias
• information bias
• confounding bias

• When study population is representative of target population.  
• Maximize by decreased bias, sampling of target population, allocation to group

• Ability to make inferences to populations beyond the target population (can results be generalized to other situations/animals? Or only applicable to target pop?  
• CANNOT be externally valid unless internally valid

• Results from an "error" in selecting individuals for a study (study pop NOT REPRESENTATIVE of target pop)
• begins before study occurs
• prevent: minimize none-response to survey, proper selection of control/comparison (inclusion/exclusion criteria), case and exposure definitions

• Bias in measurement or gathering of data (different "quality" of data between groups). (Biased observer, recall bias, measurement (accurate? Precise?), misclassification (non-differential = same frequency of error between groups, differential = error NOT the same between groups.
• minimizing: standardize protocol, accurate, precise tests, objective outcomes (not subjective), high Sp and Se in tests, Blinding.

• confound: to mix up something with something else so the individual elements are difficult to distinguish.  
• Confounding factor is associated with BOTH exposure and outcome.  Confounding bias is when this is NOT ACCOUNTED FOR in design/analysis
• Minimize: restrict study to one level of confounded, account for potential confounders, randomize clinical trials, match observational trials on confounder.

• non-differential is when frequency of classification error is the SAME between groups.  
• Differential is when frequency of classification error is different between group, differs based on disease status.

• lag
• momentum
• bias
• dispersion

when data is only available about an event of interest after some time period has passed (like outcome of breeding)

• when a parameter of interest changes slowly in response to real underlying change in a dairy.
• Rolling averages always have this - they are a measure from each of the last 12 months. Diluted by historical data

• including animals that are not relevant in computation of a parameter.  
• Affected cows/all cows
• vs
• affected cows/cows at RISK

when the distribution of values for a parameter is such that usual reporting may miss an important feature of herd behavior.

• death
• premature culling
• treatments
• discarded milk
• lost production
• delayed conception

• ?  Have to look at all the pieces - is it better to take the loss now or fix the defect?
• Individual animal decision are always marginal.  Partial budgeting helps guide, decision trees can too

includes things affected by proposed action (associated with plan).  Compare with doing nothing.

• when outcome is uncertain, calculates value of each option so you can compare end values.  
• Need to know costs associated with each choice and probabilities of each outcome, and value of each outcome.