Transcript Interpretation of the DATA

L09
Critical Review of
Epidemiologic
Studies
FORMAT OF JOURNAL ARTICLES
Abstract – a short summary of the entire article
Introduction – the context of the study, including prior research
 Context – investigator’s motivation for the study, discussing the hypothesis and objectives
Methods – the setting, design, data collection procedures*, and analysis* (which provides the most information on the
study’s validity)
 Clearly defined and appropriately measured collection of data in order to obtain an outcome
 Depends on the study type, population and associated sample size, and level of comparability between the cases vs. controls
 Major area for bias or confounding
 Potential reduction: randomization, restriction, matching, stratification, standardization, multivariation
 Measures of reported association (e.g., relative risk, odds ratio) and statistical stability (e.g., hypothesis testing)
Results – characteristics of the study population and findings from the study
Discussion – extrapolates the results, offering an interpretation* of the findings and trends while addressing
limitations
 Interpretation of the potential risk for bias or confounding based on the collected data
Conclusion – summarizes the findings and points to possible future research
 Determination of causation (in which the presence of association does not imply causation)
References
Targets areas for criticism*
CAUSATION
Evidence-based information – identifies the strengths and weaknesses within
scientific evidence, aiding in the making of informed clinical practice decisions
Causal inference – the process of how epidemiologists determine causative
and preventative disease factors
 Steps:
 Is the result valid and true (explicitly free from bias, confounding, or random error)?
 Has the exposure actually caused the disease?
 Characteristics:
 Association: the cause and effect must be statistically dependent
 Time: the cause must proceed the effect
 Direction: an established asymmetrical relationship must exist between the cause and effect (in which
the cause leads to the effect but the effect doesn’t lead to the cause)
 Positive correlation – the presence of the cause induces the effect
 Negative correlation – the absence of the cause induces the effects
 Can be influenced by host or environmental factors in which the conditions are active (with the
intervention causes a change) or static (with an unchanging given set of conditions)
CAUSATION
Sufficient-Component Cause Model (K. J. Rothman – 1976):
Sufficient cause – a set of conditions in which the absence of one event
prevents the disease from occurring
Component cause – any one of the set of conditions which are necessary
for the completion of a sufficient cause
Necessary cause – a component cause that is a member of every
sufficient cause
SUFFICIENT
CAUSE
NECESSARY
CAUSE
*
*
* * *
COMPONENT
CAUSES
CAUSAL “GUIDELINES” FOR ESTABLISHING
CAUSATION (SIR AUSTIN BRADFORD HILL):
Temporal relationship – exposure to the factor occurs before the disease development (in which the interval length
between the exposure and disease onset is important)
 Easiest to establish in a prospective cohort study
Strength of the association – demonstrates the direct association between the exposure and disease
 Stronger associations are more likely to be causal, minimizing the effects of bias and confounding
Biological Gradient – provides additional evidence that a direct causal relationship exists between the level of
exposure and disease onset
Replication of the findings – repeatability of the observed association in different scenarios and study designs
Biologic plausibility – the existence of a biological or social model to explain the association supporting current
knowledge of biology and natural history
Consideration of alternate explanations (in the event a similar relationship is observable with other exposures or
diseases)
Experiment or cessation of exposure – a decline in the risk of disease after the reduction or elimination of the
exposure
Consistency with other knowledge – interpretation of the results coincides with other data, creating an observable
association with repeatability
Specificity of the association – a single disease stemming from a single exposure, providing additional support for
causal interference when present
SURGEON GENERAL’S GUIDELINES FOR
ESTABLISHING CAUSALITY
Uses:
Denoting distinctions between association and causation in epidemiologic
research (in which associations are observed while causation is inferred)
 All of the evidence must be considered and the criteria must be weighed against
each other in order to infer the causal relationship
Critically reading epidemiologic studies
Designing epidemiologic studies
Interpretation of results
L10
Analysis and
Interpretation of
Medical Literature
JOURNAL ARTICLE – SDL
Somatic Dysfunction and Its Association With Chronic Low Back Pain, BackSpecific Functioning, and General Health: Results From the OSTEOPATHIC Trial
– John C. Licciardone, DO, MS, MBA and Cathleen M. Kearns, BA
 http://www.jaoa.osteopathic.org/content/112/7/420.full.pdf+html
Objectives:
 Identify the strengths and limitations of medical literature through critical analysis
 Identify whether the methods applied to study subjects and collect and analyze data
are appropriate and free from error
 Interpret the research study to identify any errors conducted in the design, conduct,
and interpretation of results
 Understand the impact the results of a medical journal article have on current medical
practice
 Recall how to best apply evidence based medicine within clinical practice
COLLECTION OF DATA
1. What was the context of the study?
 Somatic dysfunction is diagnosed by the presence of any 4 TART criteria: tissue texture
abnormality, asymmetry, restriction of motion, or tenderness
2. What were the objectives of the study?
 To measure the prevalence of somatic dysfunction in patients with chronic low back pain
(LBP) and to study the associations of somatic dysfunction with LBP severity, back-specific
functioning, and general health.
3. What was the primary exposure of interest? Was this accurately measured?
 Somatic dysfunction in the lumbar, sacrum/pelvis, and pelvis/innominate regions, including
key lesions representing severe somatic dysfunction
 Yes – 15 osteopathic physicians were enlisted to conduct baseline evaluations of the
participants, using the Outpatient Osteopathic SOAP Note Form as an objective tool for
measuring and recording the diagnosis and treatment of somatic dysfunction and for
categorizing the severity of somatic dysfunction in each of 14 anatomic regions on the basis
of TART criteria.
 Baseline testing: LBP severity by a Visual Analog Scale; back-specific functioning and general health by
Roland-Morris Disability Questionnaire and Medical Outcomes Study Short Form-36 Health Survey
COLLECTION OF DATA
4. What were the primary outcomes of interest? Was this accurately measured?
 Statistical significant pairwise correlations for severe somatic dysfunctions: T10-12 with ribs, T10-12
with lumbar, lumbar with sacrum/pelvis, and sacrum/pelvis with pelvis/innominate  in any anatomic
region, severe somatic dysfunction was correlated with the overall number of key lesions
 The presence of severe somatic dysfunction in the lumbar region was associated with greater LBP severity and
greater back-specific disability.
 The presence of severe somatic dysfunction in the sacrum/pelvis region was associated with greater back-specific
disability.
 An increasing number of key lesions was associated with back-specific disability and poorer general health.
 Yes – Multiple logistic regression analyses were applied to compute the adjusted odd ratios and
95% confidence intervals along with the Spearman rank correlation coefficient (ρ). Comparisons of
the predominantly nonparametric methods were computed using the Mann-Whitney test and KruskalWallis One-way Analysis of Variance to assess further relationships within the data. The data was
stratified across a broad spectrum of characteristics (e.g., age brackets, gender, health status of
various pre-existing conditions) to further separate the data when comparing p-values against the
selected 0.05 level of statistical significance.
5. What type of study was conducted?
 Cross-sectional study nested within a randomized controlled trial
COLLECTION OF DATA
6. Describe the source of the study population, process of subject selection,
sample size, and number of controls compared to cases.
 Randomized, double-blind, sham-controlled, 2×2 factorial design of 455 adult
participants between the ages of 21-69 who reported having non-specific LBP
constantly or on most days over the past 3 months
 Exclusions: presence of a “red flag” condition; low back surgery in the past year; receipt of
worker’s compensation benefits in the past 3 months; ongoing litigation involving back
problems; medical conditions that might impede OMT or ultrasound protocol implementation;
corticoid steroid use in the past month; or clinical evidence of lumbar radiculopathy (through
specified testing)
7. Is there any potential for selection bias to have occurred? If so, how did it
occur?
 Potential need to exclude patients without somatic dysfunction in clinical trials of
OMT (since a few patients did not present with any somatic dysfunction in specific
regions during the baseline structural examination)
 Unclear if the few patients without somatic dysfunction at the baseline would remain without
somatic dysfunction during the entire course of the clinical trial
COLLECTION OF DATA
8. Is there any potential for bias in the collection of information? If so,
how did it occur?
 Potential limitation involving the possible interexaminer variability in
diagnosing somatic dysfunction by using the musculoskeletal table of the
OOSOAPNF, creating the potential for overlap between scoring
 No formal assessment of provider performance or interexaminer reliability was made
although fidelity training for the OMT physicians was provided.
9. What provisions were made to minimize the influence of confounding
factors prior to the analysis of the data? Should other provisions
have been made?
 The data was stratified and compared across multiple logistic regression
analyses to minimize the potential of confounding factors.
 Yes - To further prevent possible confluence between the variables, a greater number of
participants should be evaluated in order to accurately assess all aspects of somatic
dysfunction and its possible to other variables evaluated in the study (even though no
consistent statistical significance was found).
ANALYSIS OF DATA
1. What methods were used to control confounding bias during data analysis? Were these
methods sufficient?
 Yes – Predominantly relied on nonparametric methods for analysis before dichotomizing the severity
of somatic dysfunction by combining the 3 lowest levels (none, mild, and moderate) and contrasting
these with the highest level (severe), representing clinically significant, key lesions (which maintain a
dysfunctional pattern that includes secondary dysfunctions)  multiple logistic regression analyses to
compute odds ratios and 95% confidence intervals across the stratified data
2. What measures of association were reported in this study?
 Spearman rank correlation coefficient (ρ) for severe somatic dysfunction in each anatomic region and
the overall methods
 Mann-Whitney test to compare LMP severity, back-specific functioning, and general health of
patients with and without severe somatic dysfunction in each anatomic area
 Kruskal-Wallis One-way Analysis of Variance by ranks to further assess the relationships between the
number of key lesions and LBP severity, back-specific functioning, and general health
3. What measures of statistical stability were reported in this study? How you do interpret these
measures?
 Hypothesis testing conducted at the 0.05 level of statistical significance - To test for rejection of the
H0: no association between non-specific LBP and somatic dysfunction within each anatomic area (pvalue > 0.05)
INTERPRETATION OF THE DATA
1. What were the major results of the study?
 Statistical significant pairwise correlations for severe somatic dysfunctions: T10-12 with ribs, T10-12 with lumbar,
lumbar with sacrum/pelvis, and sacrum/pelvis with pelvis/innominate  in any anatomic region, severe somatic
dysfunction was correlated with the overall number of key lesions
 The presence of severe somatic dysfunction in the lumbar region was associated with greater LBP severity and greater back-specific
disability.
 The presence of severe somatic dysfunction in the sacrum/pelvis region was associated with greater back-specific disability.
 An increasing number of key lesions was associated with back-specific disability and poorer general health.
2. How is the interpretation of these results affected by information bias, selection bias, and confounding?
Discuss the magnitude and direction of the bias?
 Possibility of inaccurate association between the specific anatomic regions mentioned and the level of somatic
dysfunction by either over- or underestimating the effect of one of the other classified factors (e.g., key lesion or
general health)
3. How is the interpretation of these results affected by non-differential misclassification? Discuss the
magnitude and direction of the misclassification?
 Misclassification bias during the baseline reporting (by the 15 osteopathic physicians) could potentially skew the
computed p-values that were numerically close to the selected 0.05 level of statistical significance (e.g., age
distributions more commonly affected by LBP and associated somatic dysfunction), overestimating the correlation
between non-specific LBP and somatic dysfunction
INTERPRETATION OF THE DATA
4. Did the discussion section adequately address the limitation of the study?
 Yes – Many of the potential limitations were extensively covered.
5. What were the author’s main conclusions? Were they justified by the
findings?
 The present study demonstrates that somatic dysfunction, particularly in the lumbar
and sacrum/pelvis regions, is common in patients with chronic LBP. Severe somatic
dysfunction in the lumbar region is directly associated with LBP severity and backspecific disability. Severe somatic dysfunction in the sacrum/pelvis region is directly
associated with back-specific disability and is inversely associated with general
health. An increasing number of key lesions were associated with greater backspecific disability and poorer general health.
 Yes – Each of the conclusions coincided with the critical p-values and rejection of the H0.
6. To what larger population can the results of this study be generalized?
 Other sufferers of non-specific LBP with similar parameters can seek treatment using
OMT techniques to relieve their associated somatic dysfunction.
L11
Outbreak
Investigation
OUTBREAKS
Epidemic – the occurrence of more disease cases than expected in a given area or among a
specific group of people over a particular period of time
 Pandemic – an epidemic occurring over a widespread area, typically affecting a substantial proportion
of the population
 Outbreak – an epidemic limited to a localized increase in the incidence of disease
Cluster – an aggregation of cases in a given area over a particular period of time without
regard to whether the number of cases is more than expected (with no inclination of chance)
Reasons for outbreaks:
 Agent: increases in the amount of virulence and the introduction into a previously unsettled area that
allows it to thrive
 Host: a change in susceptibility and influence by factors that increase possible exposure
 Environment: increased interaction between host and agent
 Enhanced mode of transmission
Main reasons for investigation: identify possible causes for prevention and control
Challenges: credibility of data sources, limited participation, specimen collection, publicity
STEPS OF AN OUTBREAK INVESTIGATION
1. Establish existence of an outbreak by determining if the observed numbers exceed the
expected levels
 Artifactual causes for increases or decreases of reporting cases: changes in reporting practices,
changes in case definition, availability of new diagnostic tests, recent media coverage
2. Verify the diagnosis by confirming the clinical diagnosis with laboratory techniques
3. Define a case based on its standard elements (e.g., clinical information, time, place, affected
individuals) and varying degrees of certainty (with associated risk factors)
 Can vary depending on the purpose
4. Identify additional cases linked by similarities to the case definition
5. Perform descriptive epidemiology by orienting the data through graphing
 Time: plotting a graph that illustrates the number of cases (y-axis) over the time of disease onset (xaxis) using an appropriate interval
 Place: geographic distribution of cases to identify possible sources and modes of transmission
 Individual: examining case characteristics based on personal exposures to establish relationships
 Determines additional information: the size of the epidemic, outliers, time course, the pattern of
spread and associated progression
STEPS OF AN OUTBREAK INVESTIGATION
6. Develop a hypothesis using descriptive epidemiology (e.g., person, place, and time with the
clinical and laboratory findings) and test the new hypothesis using analytic epidemiology
(e.g., retrospective cohort or case-control study)
7. Reconsider hypothesis by “squaring” the hypothesis to the clinical, laboratory, and
epidemiologic facts
 Development of a new hypothesis for re-testing may occur
8. Perform additional studies (if needed) by better defining the extent of the epidemic,
evaluating new laboratory methods and case-finding techniques (for improved sensitivity and
specificity), or conducting an environmental investigation
9. Implement control measures to prevent exposure and associated infection, disease, and
possible death
10. Communicate findings to the media and community for raised awareness
 Risk communication – effectively provide information about the expected type and magnitude of
an outcome from associated behaviors or exposures (in order to empower decision-making)
 Principles: prevent fear by emphasizing the establishment of a process for management of the epidemic;
acknowledge uncertainty and answer questions while being regretful of the associated events; be a role model and
assign tasks
L14
Public Health
Databases,
Resources, and
Disease Surveillance
SURVEILLANCE
Surveillance - the ongoing systematic collection, analysis, and interpretation
of health data that is essential to the planning, implementation, and evaluation
of public health practice (which is closely integrated with the timely
dissemination to those who need to know)
 Passive – regularly reporting of cases based on a standard case definition of each
particular disease (e.g., death certificates)
 Active – initiation of information collection from local or state health departments in
order to achieve more complete and accurate reporting
 Syndromic – the ongoing, systematic collection, analysis, interpretation, and
application of real-time indicators for disease, allowing for detection before public
health authorities would otherwise identify them
Uses: estimate the magnitude of the problem, determine the geographic
distribution of an illness, portray the natural history of a disease, detect
epidemics and define possible problems, generate hypotheses to stimulate
research, evaluate control measures, monitor changes in infectious agents,
detect changes in health practices, and facilitate planning
SOURCES OF DATA
Birth data – used for the calculation of community health indicators (through
the National Health Statistics System)
 Birth outcomes (e.g., delivery methods, delivery complications, birth weight)
 Demographic and socioeconomic data of the parents (e.g., race, education level)
 Reproductive history of the mother
 Prenatal care record
 Medical risk factors (e.g., gestational diabetes, previous miscarriages)
Family practices data – gathers national data on marriage/divorce, family
planning, and infertility in both men and women
Mortality data – used for the calculation of many population-based measures
(through the National Vital Statistics System)
 Demographic and socioeconomic data (e.g., age, gender, employment status and type)
 Time, place, and manner of death
 Causes of death and contributing factors
SOURCES OF DATA
Morbidity data:
 Centralized Cancer Registries – collects tumor type and associated stages of malignancy (with possible exposures)
for cancer rates
 National Notifiable Disease Surveillance System – collects exposure information and disease symptomology with
associated dates for infectious disease rates
 National Health Interview Survey – self-reporting telephone interview for the reporting of chronic disease and
medical risk factors
 Associated behavioral risk factors:
 National Health and Nutrition Examination Survey (NHANES) – gather information on the health and diet of the population through
interviews and health tests
 Behavioral Risk Factor Surveillance System – telephone interview for information on exercise and associated weight management,
nutrition intake, smoking and drinking, preventative care, and mental health
 Youth Behavioral Risk Factor Surveillance System – form-based survey administered in schools for information on sexual activity, smoking
and drinking, domestic violence, risk of suicide, and personal safety
 National Health Care Surveys – a set of surveys about the use and quality of health care and the impact of medical technology in a
variety of settings (including hospital inpatient and outpatient departments, emergency rooms, hospices, home health agencies, and
physician’s offices)
Demographic and socioeconomic data:
 US Census Bureau – administration of a decennial census and annual American Community Survey for the
enumeration and estimation of the population in order to provide the basis for population-based rates (e.g.,
poverty levels, health insurance, employment/disability, family structure)
NOTIFIABLE DISEASE SURVEILLANCE
National Notifiable Disease Surveillance System (NNDSS) – a
foundation for the state and local application of the reportable
infectious and noninfectious diseases, voluntarily passing reports
from the local and state health departments (as the role of the
Council of State and Territorial Epidemiologists (CSTE)) to the CDC
Collects a list of disease and laboratory findings of public health interest
for the creation of case definitions before disseminating the surveillance
data through the Morbidity and Mortality Weekly Report (MMWR) and
Annual Summary of Notifiable Diseases
 Case definition – uniform criteria for reporting cases
COMMUNITY HEALTH ASSESSMENT
Community Health Assessment – produces information about the health status and
needs of the community via the ongoing and systematic process of data collection,
data analysis, interpretation of results, and the distribution of findings
Purpose:
 To help inform stakeholders in the community health for decision-making (e.g., planning and
implementing of interventions, setting priorities, allocating resources)
 To document the need for resources and to bolster the community’s commitment and political
will to intervene
Functions:
 Compare the rates from similar localities to the state and national rates
 Compare the benchmark rates or Health Plan 2020 targets to the local reports
 Consider the demographic and socioeconomic comparability of the populations from which the
comparison rates were derived
 Examine both recent and trending data
 Examine subpopulation rates to reveal local issues that may be masked on a larger scale
LIMITATIONS OF SURVEILLANCE
Limitations:
Reasons for the failure to report:
 Incomplete or overwhelming volumes
of data from various sources
 Uneven application of information
technology
 Timeliness of reporting
 Fallible completeness due to
unreported cases and incomplete
reports
 Lack of awareness of legal
requirement
 Lack of knowledge of which
conditions are reportable
 Lack of knowledge of how or whom
to report
 Assumption that someone else will
report the case
 Intentional failure to report to
protect patient privacy
 Insufficient reward for reporting
 Insufficient penalty for not reporting
L15 – L17
Biostatistics
POPULATION
Population – all items in a study or group
Sample – a subset of the population (as a representative subgroup)
 Simple random – every member of the population has an equal chance of
being selected
 Systemic sampling – samples are selected over a fixed pattern or time interval
 Stratified – when known differences (categories) exist in a population, samples
can be taken from each category that are proportional to their volume in the
total population
 Cluster – only used when a population is known to be relatively unvarying
 Convenience – selects the most readily available members of a population
 Quota – takes a percentage of the population
Uses of sampling: time limitation, cost of the study, representation of
the population
DESCRIPTIVE STATISTICS
Descriptive statistics – simple graphical numerical techniques to summarize information about the data
Statistical inference – how the information contained in a sample can be used to draw conclusions about
the population
Random variables – anything capable of being measured
 Measures of centers – represents the location of data
 Mean (μ) – the calculated average from the sum of the observations divided by the number of observations
 Median – the middle number in a set of numbers ordered from smallest to largest
 Even number of data? Calculate the mean of the two center values
 Mode – the most frequently occurring number in a set of numbers
 Measures of dispersion – represents the variability of the data observed as the spread of the distribution
 Standard deviation – a measure of the variation in a set of data away from its mean
 Standard error – estimates the probable error (or variation) of the sample mean against the estimated population mean
𝒔
s = standard deviation
SE𝒙 =
n = sample size
𝒏
 95% confidence intervals – defines a range of values likely to contain the true population mean 𝟗𝟓% C.I. = μ ± 𝟐 SE
 What does this mean? We are 95% confident that the true mean value in the population should be between (μ - SE) and (μ + SE).
 Range – difference between the minimum and maximum value
 Quartile – splitting of the data into four evenly subdivided categories around the median (in intervals of 25%)
 Minimum – lowest value in a set of data
 Maximum – highest value in a set of data
MEASURES OF DISPERSION
Standard deviation:
Example: The average on the biostatics exam is 85 with a standard
deviation of 5. Using the understood standard deviation percentiles,
calculate the standard deviation ranges.
 68% of the class: (85-5) and (85+5) thus 68% of the students received an 80-90
 95% of the class: (85-(2×5)) and (85+(2×5)) thus 95% of the students received a 7595
 99.7% of the class: (85-(3×5)) and (85+(3×5)) thus 99.7% of the students received a
70-100
SUMMARIZING DATA
Categorical random variables –
data capable of being counted
and recorded (as a frequency or
percentage)
Continuous random variables –
data with no finite end (through
the use of summary statistics)
 Summary statistics (e.g., mean and
associated standard deviations,
minimums and maximums, median
and mode)
l
DATA
Quantitative data – used to determine the relationship between an independent variable (the
predictor variable) and a dependent variable (the outcome variable) in a population
 Descriptive – subjects are measured once, establishing associations between the variables
 Experimental – subjects are measured before and after the treatment
Qualitative data – involves an in-depth understanding of human behavior and the reasons that
govern it, relying on the reasons behind the various aspects of behavior (used as a research
methodology in the social sciences)
Categorical data:
 Nominal data – counted data (that is not measured on a scale) with no ranking order (e.g., blood
groups)
 Ordinal data – ordered data with 2+ categories of classification (which can be used to create a
ranking order) (e.g., severity of illness)
 Dichotomous (binary) data – data that consists of counting in whole numbers (with favourably implied
direction) (e.g., healthy vs. sick)
Continuous (dimensional) data – measured data that can take on any value along a
continuous scale (with no finite value) (e.g., birth weight)
CONTINUOUS DATA
Normal distribution:
 Data is continuous
 The mean, median, and mode are equal to
each other, existing as unimodal data
 A certain percentage of the data falls
within a specific standard deviation (68%,
95%, and 99.7%)
Left skewed distribution – the mean falls
to the left of the median and mode
j
Right skewed distribution – the mean falls
to the right of the median and mode
Mean
Median
Mode
Mode
Median
Mean
HYPOTHESIS TESTING
Hypothesis testing – a specified claim or theory used to compare treatments, acting
as a scientific inquiry into the connection between cause and effect
 Uses the scientific method to test the validity of the hypothesis based on observed data
Goal: to see if there is sufficient statistical evidence to reject a presumed null
hypothesis in favor of an alternate hypothesis
 Null hypothesis (H0) – the claim or theory currently presumed to be true in which there is no
true underlying difference between the groups
 Alternate hypothesis (H1) – the claim or theory to be proved in which there is a difference
between the groups
Steps:
1. Develop the null and alternate hypotheses
2. Establish an appropriate α-level
3. Perform a suitable test of statistical significance on the appropriately collected data
4. Compare the p-value from the test with the established α-level
5. Conclude the result by either rejecting the null hypothesis in favor of the alternative (pvalue < α) or failing to reject the null hypothesis (p-value > α)
STATISTICAL SIGNIFICANCE
Statistical significance:
P-value > α? There is no statistically
significant difference.
P-value < α? There is a statistically
significant difference.
Type I error (α) – a difference exists
when there really is not a difference
Type II error (β) – a difference does not
exist where there really is a difference
H0 IS TRUE
H0 IS FALSE
REJECT H0
TYPE I
ERROR
CORRECT
FAIL TO
REJECT H0
CORRECT
TYPE II
ERROR
POWER ANALYSIS
Power (1-β) – the chance of finding a significant effect when one does exist
 If the null hypothesis is false, what is the probability that the data from the experiment
will reject the null hypothesis?
 Goal: to stroke a balance among the balance in order to achieve the most sensitive
test given with the available resources
Factors affecting power:
 Effect size – the minimum signal that is required for detection (in which the larger the
size of the effect, the better the chances are of finding it)
 Significance level – direct relationship between α and β (and subsequently α and
power)
 Appropriate sample size (in which as the sample size increase so does the power of
the associated test)
 Population standard deviation – variability in the process which can obscure the
signal, rendering the test less powerful (in which greater data variability requires a
larger sample size to compensate)
PARAMETRIC STATISTICS
Parametric statistics – assuming the distributions of the variables being assessed belong to the
known parameterized families of probability distributions
Student T-test – comparison between independent samples (in
which one sample in no way affects the other sample) in order
to evaluate the difference in means between the groups
 Mann-Whitney U-test – comparison between independent samples of ordinal (ranked) data, combining
the groups to rank the entire data set (e.g., when data is not normally distributed)
 H0: the populations are from the same data set
 H1: one population is larger than the other
 Paired T-test – comparison within dependent paired (or matched) samples (in which an observation from
one sample determines an observation taken from the other sample), evaluating the differences in
means within groups
Analysis of Variance (ANOVA) – comparison of independent
samples with continuous data, evaluating the differences in 2+
categorical groups, (in which additional testing can be done to
ascertain where (if any) the differences are)
 Kruskal-Wallis test (One-way ANOVA) – comparison of independent samples from 2+ groups to test
whether the continuous data samples are from the same distribution
NON-PARAMETRIC STATISTICS
Non-parametric statistics – used when nothing is known about the parameters of the
variable of interest in the population (e.g., Wilcoxon Signed-Rank Test, Pearson Χ2)
Wilcoxon Signed Rank test – comparison within dependent samples (as an
alternative to the Paired T-test) in which there are repeated measurements of the
data (but it is not normally distributed), assessing whether the population mean ranks
differ
NO
Chi-Square Test (Χ2) – comparison of the proportions for
DISEASE
TOTAL
DISEASE
treatments and outcomes using a ratio of actual-toexpected, using a 2×2 contingency table, based on the
EXPOSURE
a
c
m
H0 assumption that the expected values from the
classifying values (factors) are true
 Fisher’s Exact test – comparison of the proportions for
treatments and outcomes using a ratio of actual-to-expected
when the sample size is small, using a 2×2 contingency table,
based on the H0 assumption that the expected values from
the classifying values (factors) are true
NO
EXPOSURE
b
d
n
TOTAL
r
s
N
CORRELATION
Correlation – provides a visual display of the relationship between 2
variables for prediction, showing how one variable directly or
indirectly increases or decreases with another variable
Interpretation of the correlation statistic (in which correlation does not
equal causation due to confounding variables):
WEAK CORRELATION: values between 0.0 – 0.3
MODERATE CORRELATION: values between 0.31 – 0.7
STRONG CORRELATION: values > 0.7
i
NO CORRELATION
STRONG-POSITIVE
CORRELATION
STRONG-NEGATIVE
CORRELATION
EXACT LINEAR
CORRELATION
PRESENCE OF
AN OUTLIER
CORRELATION
Spearman Rank Correlation Coefficient (ρ) – assesses the
relationship between 2 continuous variables (in which at least
one is not normally distributed) in a single sample
Kendal Rank Correlation Coefficient (Τ) – measures the ranked
correlation between 2 variables from smaller sample sizes (in
which the order of the data are ranked by quantity)
L18 – L19
Applied Biostats
RACIAL AND ETHNIC DISPARITY IN LOW
BIRTH WEIGHT IN WAYNE COUNTY, NC
Abstract: Low birth weight is a leading cause of infant mortality.
Unfortunately, despite declining rates of infant mortality, racial
and ethnic disparities in both low birth weight and infant mortality
rates persist. In this teaching case, a clinical vignette is used to
draw attention to this public health priority in Wayne County, NC.
Students learn essential epidemiological skills, such as identifying
limitations of sources of data and calculating relative risks, using
the example of low birth weight. In performing these skills,
students also identify etiologies for such disparity. Finally, students
discuss interventions that, when implemented, may decrease infant
mortality rates.
L20
Evidence-based
Prevention: USPSTF
USPSTF
United Sates Preventative Services Task Force (USPSTF) – the leading
independent panel of private-sector experts in primary care and prevention
sponsored by the Agency for Healthcare Research (AHRQ)
 Mission: to improve the quality, safety, efficiency, and effectiveness of health care for
all Americans
 Framework: screening for early detection and treatment of at-risk individuals decreases morbidity
and/or mortality
 Functions: conducts rigorous and impartial assessments on the scientific evidence
covering effective preventative services (e.g., screening, counseling, preventative
medications), creating the “gold standard” recommendations (for asymptomatic
children and adults) that should be routinely incorporated into clinical preventative
services
 Does not conduct the research—only evaluates the benefits and harms of existing evidence
 Topic nominations: suggestions for new preventative services from the public; existing topic
reconsiderations due to the availability of new evidence, changes in the public health burden of the
condition, or availability of new screening tests with new evidence
 Members: 16 volunteers represented by a chair and vice-chair that serve 4 year terms
(as appointed by the AHRQ Director)
RECOMMENDATION GRADES
Levels of certainty regarding net benefits:
 High: the available evidence from well-designed and well-conducted representative studies is consistent
 unlikely to be strongly affected by the results of future studies
 Moderate: the available evidence is sufficient but confidence in the preventative service effects is limited
by the quality of the study  the magnitude or direction of the observed preventative effect could
change as more information is made readily available
 Low: the available evidence is inefficient to assess the effects on the preventative health outcomes due to
severe limitations of the study  requires more information for a better estimation of the effects on the
preventative health outcomes
Grade
Definition
A
The USPSTF recommends the service. There is high certainty that the net benefit is substantial.
B
The USPSTF recommends the service. There is high certainty that the net benefit is moderate,
or there is moderate certainty that the net benefit is moderate to substantial.
C
Clinicians may provide this service to selected patients depending on individual circumstances. However, for most individuals
without signs or symptoms there is likely to be only a small benefit from this service.
D
The USPSTF recommends against the service. There is moderate or high certainty that the
service has no net benefit or that the harms outweigh the benefits.
;
I Statement
The USPSTF concludes that current evidence is insufficient to assess the balance of benefits & harms of the service.
PUBLIC HEALTH BURDEN
Tobacco use is the leading preventable
cause of death in the U. S., resulting in
400,000 deaths annually
 USPSTF recommendations:
 Clinicians ask all adults about tobacco use and
provide tobacco cessation interventions for those
who use tobacco products – A
 Clinicians ask all pregnant women about tobacco
use and provide augmented, pregnancy-tailored
counseling for those who smoke – A
 Tobacco counseling:
1. Ask about tobacco use
2. Advise to quit through clear personalized
messages
3. Assess willingness to quit
4. Assist to quit
5. Arrange to follow-up and support
Alcohol misuse is the 3rd leading cause
of preventable death in the U. S.,
causing 85,000 deaths annually
 USPSTF recommendations:
 Screen adults aged 18 years or older for
alcohol misuse and provide persons engaged in
risky or hazardous drinking with brief
behavioral counseling interventions to reduce
alcohol misuse – B
 Current evidence is insufficient to assess the
balance of benefits and harms of screening and
behavioral counseling interventions in primary
care settings to reduce alcohol misuse in
adolescents – I
 Screening for alcohol misuse: AUDIT
(Alcohol Use Disorders Identification Test),
Abbreviated AUDIT-C (Alcohol Use
Disorders Identification Test - Consumption),
Single question screening
PUBLIC HEALTH BURDEN
Illicit drug use is ranked among the top 10 preventable risk factors for years
of healthy life lost
 USPSTF recommendation: current evidence is insufficient to assess the balance of
benefits and harms of screening adolescents, adults, and pregnant women for illicit
drug use – I
Heart disease is the leading cause of death in the U. S. (which can be reduced
by healthy diet and exercise)
 USPSTF recommendation (for the general adult population without a known
associated diagnosis): since the evidence indicating the health benefit of counseling is
small, clinicians may choose to selectively provide preventative services based on risk
factors, patient readiness for change, social support, community resources, etc. – C
Approximately 69.2% of adult men and women are overweight (with 35.9%
being obese), contributing to known associated health risks
 USPSTF recommendation: screening all adults for obesity such that clinicians offer or
refer patients with a body mass index (BMI) of 30 kg/m2 or higher to intensive,
multicomponent behavioral interventions – B