Biostatistics Biostatistics is statistics applied to biology Design

Biostatistics • Biostatistics is statistics applied to biology • Design of experiments • The limitations when working with human subjects • Non-normality

Biostatistics text topics Bland, Ch. 2 A short historical review Design of experiments Bland, Ch. 13 Contingency tables Dawson and Trapp, Ch. 9 Survival analysis Logistic regression ROC analysis

Proofs • Mathematical proofs: • Emperical proofs: F = m*a

Medical research • Experimental studies – Clinical trials • Observational studies – Cohort study • Longitudinal, and prospective, time and patient consuming – Case-control study • Can be applied with low sampling number, difficult to choose the control – Cross-sectional study / survey • Inexpensive, historical, provides the current – Case-series study • Usually reports unexpected clinical observations • Meta-analysis • Reviews

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Concurrent treatment • The wonders of the CT scanner – Patients treated in 1978 for stroke scanned by CT scanner – Patched paired with stroke patients from 1974 (before the scanner) C-T scan in 1978 Pairs with 1978 better than 1974 9 (31%) Pairs with same outscome 18 (62%) Pairs with 1978 worse than 1974 2 (7%)

Concurrent treatment • The wonders of the CT scanner – Patients treated in 1978 for stroke NOT scanned by CT scanner – Matched paired with stroke patients from 1974 (before the scanner) C-T scan in 1978 No C-T scan in 1978 Pairs with 1978 better than 1974 9 (31%) 34 (38%) Pairs with same outscome 18 (62%) 38 (43%) 2 (7%) 17 (19%) Pairs with 1978 worse than 1974

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Random allocation • Doctors chose for them self if tuberculosis should receive a BCG vaccine or be in the control group No. of children 1927 -32 Selected allocation deaths from TB # visits to the clinic cooperation BCG 445 3 (0. 67%) 3. 6 43% Control 545 18 (3. 30%) 1. 7 24%

Random allocation • Tuberculosis patients were allocated randomly to receive BCG vaccine or to be in the control group No. of children deaths from TB # visits to the clinic cooperation 1927 -32 Selected allocation BCG 445 3 (0. 67%) 3. 6 43% Control 545 18 (3. 30%) 1. 7 24% 1933 -44 Random allocation BCG 566 8 (1. 41%) 2. 8 40% Control 528 8 (1. 52%) 2. 4 34%

Random allocation • OK: – Toss a coin – Throw dices – Computer programs – http: //www. random. org/ • Not OK – Alternation – By date

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Investigator Bias • Do not allocate by date of arrival: – Patients arriving at even days should be treated, and patients arriving at odd days should act as control.

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Volunteer Bias • People volunteering for experiments differs form the general population – Obviously more compliant – Volunteers for experiments at AAU?

Volunteer Bias • Salk poliomyelitis with two different designs: – Randomized controls 2 nd grades were invited to participate and randomized to either vaccine or saline injection – Observed controls 2 nd graters was offered vaccine. 1 st and 3 rd graders were unvaccinated controls

Volunteer Bias Randomized control Observed control Study group # in group Number of cases (per 100000) Vaccinated 200745 33 (16) control 201229 115 (57) Not inoculated 338778 121 (36) Vaccinated 2 nd grade 221998 38 (17) Control 1. and 3. grade 725173 330 (46) Unvaccinated 2. grade 123605 43 (35)

Volunteer Bias Randomized control Observed control Study group # in group Number of cases (per 100000) Vaccinated 200745 33 (16) control 201229 115 (57) Not inoculated 338778 121 (36) Vaccinated 2 nd grade 221998 38 (17) Control 1. and 3. grade 725173 330 (46) Unvaccinated 2. grade 123605 43 (35) • If vaccine was as ineffective as saline we would expect 200. 745*57/100. 000 = 114 cases. • In the entire randomized control we would expect 114+121 = 350 cases or 46 pr. 100. 00 subjects.

Selection of subjects • Low variability makes it easier to detect differences in the treatments • Inclusion and exclusion criterions – Acute bilateral pulmonary tuberculosis – Bacteriologically proved – Between 15 and 30 years – Unsuitable for other treatment • Narrow inclusion criterion makes it difficult to conclude on the general population

Prevalence What if patients die before inclusion?

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Intention to treat Randomized control Observed control Study group # in group Number of cases (per 100000) Vaccinated 200745 33 (16) control 201229 115 (57) Not inoculated 338778 121 (36) Vaccinated 2 nd grade 221998 38 (17) Control 1. and 3. grade 725173 330 (46) Unvaccinated 2. grade 123605 43 (35) • Analyzing the data as we had intention to treat the subjects although they may refuse treatment. • All 2 nd graders in the observed control study were intended to be vaccinated: (38+43)/(221998+123605)=23 • Conservative estimate

Subject flow

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Cross-over design Patient number # attack on placebo Pronethaol 1 71 29 2 323 348 3 8 1 4 14 7 5 23 16 6 34 25 7 79 65 8 60 41 9 2 0 10 3 0 11 17 15 12 7 2

Cross-over design Patient number # attack on difference placebo Pronethaol Placebo-Pronetahol 1 71 29 42 + 2 323 348 -25 - 3 8 1 7 + 4 14 7 7 + 5 23 16 7 + 6 34 25 9 + 7 79 65 14 + 8 60 41 19 + 9 2 0 2 + 10 3 + 11 17 15 2 + 12 7 2 5 +

Cross-over design • • • Each subject acts as his/hers own control Randomization Carry-over effect of treatment or testing Treating ‘symptoms’ Better than case-control (The healthy worker)

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Placebo (response bias) • Any treatment may help • The subject should be blinded to the kind of treatment • Effective treatment should be better than a sham treatment

Placebo • Take the red pill

Double blind studies • When the assessment is made the investigator should be blind to the treatment. • Can all studies be blinded? • Communication between subjects

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid

Ethical considerations • The Declaration of Helsinki – Reduce suffering of the subjects • Local ethics comity – Den videnskabsetiske kommité consists of both experts and laymen. – Secures that the rights of the subjects are being followed, especially as to the information given to the subjects.

Ethical considerations • Informed Consent – Subjects must be informed about the nature, purpose, risks etc. in the study. – The subjects should give their consent before being enrolled and have the right to withdraw from the study at any time.

Stanley Milgrams "watershed" experiment

Comparing Treatments • The treatments must be: – – – – – Applied concurrently Treated by the same investigator Random allocation into treatment groups Avoid cheating (investigator bias) Volunteer bias Intention to treat Use cross-over design if possible Double blinded Ethically solid