More Terminology

1
More Terminology

• Causation
• Statistical

2
Problems with data collection

• Unreliable sources
• True even for statistics reported by health
  authorities or government agencies
• Improper use of data ranges
• Where a rate is reported as a range, such as 3
  to 5 million people, the lower or higher number
  is arbitrarily chosen
• May cause some estimates to be lower or higher

3
Problems with data cont.
Extrapolation - the process of inferring unknown
information from known data.

• Improper extrapolation
• Estimates from a small sample used to describe a
  larger population group
• Has various inherent limitations to its accuracy
• Different age groups
• Data may refer to a particular age group that may
  not reflect the overall prevalence in the entire
  population of all ages

4
Statistical Tests

• Employed in explanatory studies
• Assess the role of chance as explanation of
  pattern observed in data
• Most commonly assesses how 2 groups compare on an
  outcome
• Is the pattern most likely not due to chance?
• The difference is statistically significant
• Is the pattern likely due to chance?
• The difference is not statistically significant
• No matter how well the study is performed, either
  conclusion could be wrong

5
Descriptive Statistics

• Used to describe what is (or was)
• Patient characteristics at baseline
• Examples mean, standard deviation, median,
  range, percentage
• Assess group comparability on baseline
  characteristics
• Assess generalizability of results to target
  population

6
How would you describe these data?

7
Analytical (inferential) statistics

• Assess statistical significance with confidence
  intervals and p-values
• Within and between group differences
• Make inference about target population
• Must be appropriately interpreted in the context
  of the research question and the study design
• Significance when question or design is bad?

8
Difficulties in the process of drawing inference

• Exposure of healthy subjects to suspected agents
  ethical?
• Withholding suspected beneficial agents ethical?
• Thus, epidemiologic evidence from observational
  studies very valuable

9
Bradford Hill Criteria for assessing causality

• Were developed for use in the field of
  occupational medicine, but  have been widely
  applied in other fields
• Criteria serve as a general guide, and are not
  meant to be an inflexible list
• Findings of experimental research is preferred,
  but oftentimes research questions and variables
  do not lend themselves to rigorous investigation

10
Bradford Hill Criteria cont.
Hill AB The environment and disease association
or causation?Proceedings of the Royal Society of
Medicine 1965, 58295-300.

• Strength of Association
• The stronger the relationship between the
  variables, the less likely it is that the
  relationship is due to something extraneous
• Temporality
• It is logically necessary for a cause to precede
  an effect in time

11
Bradford Hill Criteria cont.

• Consistency
• Multiple observations, of an association, with
  different people under different circumstances
  and with different measurement instruments
• Theoretical Plausibility
• It is easier to accept an association as causal
  when there is a rational and theoretical basis
  for such a conclusion

12
Cagnie B, et al. Changes In Cerebellar Blood Flow
After Manipulation Of The Cervical Spine Using
Technetium 99mEthyl Cysteinate Dimer. J
Manipulative Physiol Ther 200528103- 107.
13
Bradford Hill cont.

• Coherence
• The association must be coherent with other
  knowledge
• No plausible competing theories or rival
  hypotheses
• Specificity in the causes
• In the ideal situation, the effect has only one
  cause
• Showing that an outcome is best predicted by one
  primary factor adds credibility to a causal claim

14
Bradford Hill cont.

• Dose Response Relationship
• There should be a direct relationship between the
  risk factor and status of the disease
• Experimental Evidence
• Related research that is based on experiments
  will make a causal inference more plausible

15
Bradford Hill cont.

• Analogy
• Sometimes a commonly accepted phenomenon in one
  area can be applied to another area
• The key criteria required to establish causation
  are commonly
• 1) temporal relationship, 2) specificity, 3)
  biological plausibility, 4) coherence

16
Bradford Hill cont.

• When using them, dont forget Hills advice
• None of these nine viewpoints can bring
  indisputable evidence for or against a cause and
  effect hypothesis . What they can do, with
  greater or less strength, is to help answer the
  fundamental questionis there any other way of
  explaining the set of facts before us, is there
  any other answer equally, or more, likely than
  cause and effect? (Cited in Doll, 1991)

17
Smoking causes lung cancer

• Where is the RCT that gave evidence to this?
• Decades of observational research
• Observing the same thing in a wide variety of
  settings
• Some free of some types of bias, others free of
  other types of bias

18
Smoking lung cancer

• It takes a lot of observational data to even
  begin to suggest causation!
• Strength of association Consistency
  Specificity Temporality Biologic gradient
  (dose/response) Biologic plausibility
  Coherence of evidence

19
Smoking lung cancer Example of causality
guidelines

• Cohort studies clearly demonstrate that smoking
  precedes lung cancer
• Risk ratios between smoking and lung cancer are
  high in many studies
• The more cigarette smoke inhaled over a lifetime,
  the greater the risk of cancer

20
Example of causality guidelines cont.

• Association found in both sexes, all races, all
  strata of socio-economic status (SES), etc.
• Burning tobacco produces carcinogenic compounds
  which contact pulmonary tissue

21
Bradford Hill examplesmoking and lung cancer

• Strength of Association
• The lung cancer rate for smokers is quite a bit
  higher than for nonsmokers (e.g., 80 of female
  and 90 of male lung cancer cases were smokers)
• Temporality
• Smoking precedes the onset of lung cancer in the
  vast majority of cases

22
Bradford Hill example cont.

• Consistency
• Different methods produced the same result (e.g.,
  prospective and retrospective studies)
• The relationship is apparent in different kinds
  of people (e.g., males and females)
• Theoretical Plausibility.
• The biological theory of smoking causing tissue
  damage which over time results in cancer in the
  cells is highly plausible

23
Bradford Hill example cont.

• Coherence
• The conclusion that smoking causes lung cancer
  makes sense, given the current knowledge about
  the biology and history of the disease
• Specificity in the causes
• Lung cancer is best predicted from among smokers,
  although there are other causes

24
Bradford Hill example cont.

• Dose Response Relationship
• Data shows a positive linear relationship between
  the amount smoked and the incidence of lung
  cancer
• Experimental Evidence
• Tar painted on laboratory rabbits ears produces
  cancer in the ear tissue over time
• As a result, it was clear that carcinogens were
  present in tobacco tar

25
Bradford Hill example cont.

• Analogy
• Induced smoking with laboratory rats showed a
  causal relationship
• Therefore, it was not a great jump for scientists
  to apply this to humans

26
Smoking and cirrhosis of the liver?

• The incidence of cirrhosis of the liver is
  associated with cigarette smoking. Does this mean
  smoking causes cirrhosis?
• Heavy smokers tend to be heavy drinkers, the
  statistical association is there, but in this
  case is probably a confounding variable.
  Excessive consumption of alcohol is a more likely
  cause.

27
ANNUAL PHYSICAL AND ECONOMIC COST OF MEDICAL
INTERVENTION
Death by Medicine Gary Null PhD, Carolyn Dean MD
ND, Martin Feldman MD, Debora Rasio MD. October
2003 www.NutritionInstituteOfAmerica.org
28
Key analytical statistical concepts p and r

• P-values and confidence intervals
• Reflects measure of effect relative to variation
  and sample size
• Variation
• A function of consistency of the data
• Example Mean value of 5
• 1, 10, 2, 11, 1
• 5, 4, 3, 6, 7 (less variation)

29
Pearsons correlation coefficient

• The most common measure of correlation
• AKA - Pearson Product Moment Correlation
  Coefficient - r
• The correlation between two variables reflects
  the degree to which the variables are related
• It reflects the degree of linear relationship
  between two variables
• Ranges from 1 to -1

30
Pearsons r
Line graph representing the degree of correlation
between variables X Y In this case it is
perfect
31
Positive negative r-values

• When large values of one set are associated with
  large values of the other, its a positive
  correlation
• When small values of one set associated with
  large values of the other, its a negative
  correlation
• When the values in both sets are unrelated, the
  correlation approaches zero

32
R2

• R2 is simply the square of r
• R2 represents the amount of change in one
  variable that can be explained by the change in
  another variable
• So if r .79, R2 .62
• And one can say that 62 of the change in one
  variable is explained by the change in another
  variable

33
P-values (p probability)

• A statistical value that indicates the
  probability that the observed results are due to
  chance alone
• Determines how confident we can be in a studys
  conclusion
• Typically written this result was significant at
  p ? 0.05
• 0.05 is most commonly used

34
P-value ? 0.05 - Example

• 50 patients in each treatment group
• 25 get better with tx. A
• 35 get better with tx. B
• Statistically speaking, and all other things
  being equal, we could expect this result to occur
  by chance no more than 5 times in every 100
  trials
• Statistically, it is possible, but unlikely, that
  the groups could actually be the same, yet a
  difference would be detected

35
Significant Findings

• One needs to consider BOTH statistical and
  clinical significance
• Statistical has to do with a mathematical
  analysis of the studys results
• Clinical has to do with the applicability of
  these results to clinical practice
• You can have one without the other

36
Clinical Significance

• Is it clinically important?
• If a study found a statistically significant
  difference of 2O of cervical ROM, would that much
  difference make a difference clinically?
• Defined before a study is conducted
• How much difference in ROM would doctors be
  concerned with?
• Assessed by mean improvement in outcome measure

37
Statistical Significance

• A term used to indicate the likelihood that the
  results from a study are due to chance
• Statistically significant
• If not due to chance
• Not statistically significant
• If they are due to chance

38
Statistical Significance

• How do we know whether or not the results from a
  study are due to chance?
• Statistical tests are carried out on the results
  to determine the probability that they are due to
  chance
• Typically must have ? 0.05 probability of being
  due to chance in order to be considered
  statistically significant

39
Statistical Significance

• In other words, we must be 95 confident (based
  on statistical testing) that the results are not
  due to chance in order to say that the studys
  results are statistically significant

40
Example

• A study reports that it found LB pain scores
  improved after chiropractic manipulation, and
  that these findings are statistically significant
• This means that the researchers are reasonably
  certain that pain scores did actually improve
  after chiropractic
• If the findings were not statistically
  significant, any improvement reported may have
  been due to chance, rather than a result of the
  intervention

41
Interpreting p-values

• plt0.01 ? highly statistically significant
  difference
• 0.01?p?0.05 ? statistically significant
  difference
• 0.05ltp?0.10 ? borderline statistically
  significant difference (but not significant)
• pgt0.10 ? no statistically significant difference

42
Possible scenarios

• Statistically and clinically significant findings
• Clinically significant, but not statistically
  significant
• Statistically significant, but not clinically
  significant

43
Reliability

• The degree to which a test consistently measures
  what it is supposed to measure
• Results should be consistent when tests are
  repeated several times by the same examiner or
  between 2 or more examiners

44
Validity

• The degree to which a test measures what it is
  intended to measure
• The test is valid for a particular purpose or
  group
• Validity does not directly deal with the strength
  of the conclusions

45
Reliability and Validity

• Reliability is consistency
• Lack of reliability is a problem with random
  error
• CHANCE
• Validity is TRUTH or ACCURACY
• Lack of validity is a problem with systematic
  error
• BIAS

46
Bias

• Deviation of results or inferences from the
  truth, or processes leading to such systematic
  deviation
• Any trend in the collection, analysis,
  interpretation, publication, or review of data
  that can lead to conclusions that are
  systematically different from the truth

47
Reliability and Validity

• The center of the target is the concept that you
  are trying to measure
• For each person you are measuring, you are taking
  a shot at the target