Services
0 Items
CATEGORIES
CATEGORIES MENU
Subscribe to this RSS feed

BIOSTATISTICS FOR UPGRADING CLASS (10)

1 Background

Biostatistics is a fundamental discipline in biomedical sciences.

  • Why it matters:
    • Modern medicine is increasingly quantitative (e.g., genomics, clinical trials, precision medicine).
    • Health sciences face uncertainty and variability, which must be quantified.
    • Translational research requires robust statistical methodology to bridge laboratory findings with patient care.
  • Key epistemological point: Biostatistics is not merely “number-crunching.” It is about inference—drawing valid, reproducible, and generalizable conclusions from incomplete or noisy data.

2. Definitions and Conceptual Foundations

  • Statistics (Lovitt): Collection, classification, and interpretation of numerical facts for explanation and comparison.
  • Biostatistics:

1.  Application of statistics to biology, medicine, and public health.

2.  Tools and methods for planning, conducting, and analyzing biomedical research.

  • Modern extensions:
    • Bayesian Biostatistics: Integrates prior knowledge with observed data. Widely used in adaptive clinical trials.
    • Computational Biostatistics: Use of high-dimensional data analytics, machine learning, and AI.
    • Causal Inference: Beyond correlation — establishing “what works for whom under what conditions.”

3. Historical Development

  • Pre-20th century: Descriptive statistics in medicine (e.g., mortality tables).
  • 1929: Dunn’s work linked physiology with statistics.
  • 1937: Bradford Hill’s articles codified statistical medicine.
  • 1948: First RCT on streptomycin for TB → cornerstone of evidence-based medicine.
  • 1950s–1980s: Rapid growth of medical statistics with clinical trial methodology.
  • 21st century: Integration with informatics, bioinformatics, and big data.

Pioneers:

  • R.A. Fisher – ANOVA, design of experiments, maximum likelihood.
  • Karl Pearson – correlation, chi-square test.
  • Douglas Altman – medical reporting guidelines (CONSORT).
  • C.R. Rao – advanced statistical inference, multivariate analysis.
  • Gauss – normal distribution (central to biostatistical inference).

4. Medical Uncertainty and Its Sources

Medicine is filled with uncertainties, which biostatistics aims to reduce:

  • Intrinsic Biological Variation
    • Age, sex, genetics, BMI.
    • Physiological variation (e.g., BP, HR, lab values).
    • Pathological variation (e.g., tumor heterogeneity).
  • Environmental Influences
    • Nutrition, smoking, pollution, sanitation, stress.
  • Sampling Error
    • We cannot study the “entire population,” only samples.
    • Sampling variability introduces uncertainty.
  • Measurement Error
    • Observer bias, instrument calibration, misclassification (e.g., false positives in diagnostic tests).
  • Knowledge Gaps
    • Unclear disease mechanisms.
    • Unpredictable progression (e.g., cancer survival).

 

5. Roles of Biostatistics in Health Sciences

A) Clinical Medicine

  • Defining normal vs. abnormal using reference distributions.
  • Evaluating diagnostic tests (ROC curves, sensitivity/specificity, likelihood ratios).
  • Designing clinical trials (randomization, blinding, sample size, effect size).
  • Personalized medicine: risk prediction models (e.g., Framingham Risk Score, polygenic risk scores).

B) Preventive Medicine & Public Health

  • Surveillance: quantifying disease burden.
  • Identifying risk factors (case-control, cohort studies).
  • Health program evaluation (impact of vaccination campaigns, legislation).

C) Health Systems Planning

  • Summarization of community data (prevalence, incidence, DALYs, QALYs).
  • Evaluation of interventions (cost-effectiveness, cost-utility).

D) Medical Research

  • Research design (cross-sectional, longitudinal, experimental).
  • Validation of measurement tools (reliability, construct validity, Cronbach’s alpha).
  • Analysis (classical methods + modern machine learning).

6. Data: Types and Sources

  • Qualitative (categorical): Nominal, ordinal.
  • Quantitative (numerical): Discrete, continuous.
  • Primary data: Surveys, interviews, experiments.
  • Secondary data: Hospital records, census data.

Advanced perspective:

  • Big Data Sources: Electronic health records (EHRs), genomic databases, wearable devices.
  • Challenges: Missing data, selection bias, confounding.

7. Scales of Measurement and Their Implications

  • Nominal: Categorical, no order (sex, race). → Tests: Chi-square, logistic regression.
  • Ordinal: Ordered but unequal intervals (severity scores). → Tests: Mann-Whitney U, ordinal regression.
  • Interval: Equal intervals, no true zero (temperature). → Tests: t-test, ANOVA.
  • Ratio: Equal intervals, true zero (age, weight). → All parametric methods.

PhD Note:
Misuse of scales leads to flawed inference. E.g., treating ordinal data as interval can inflate Type I error.

8. Clinimetrics – Bridging Qualitative and Quantitative

  • Purpose: Converts clinical observations into measurable scales.
  • Examples:
    • Apgar Score (neonatal outcome).
    • APACHE Score (ICU severity index).
    • Smoking Index.

Advanced uses:

  • Prognostic models (e.g., SOFA score in sepsis).
  • Composite indices in epidemiology (e.g., socioeconomic status indices).

9. Data Summarization and Visualization

  • Tables: Frequency distribution, cross-tabulation.
  • Graphs:
    • Categorical: Bar charts, pie charts.
    • Quantitative: Histogram, box plot, scatterplot.
    • Advanced: Survival curves, forest plots, heatmaps.

 

10. Statistical Workflow in Biomedical Sciences

1.  Planning – Define hypothesis, select study design.

2.  Design – Sampling, randomization, blinding.

3.  Data Collection – Minimize bias, ensure reliability.

4.  Data Processing – Cleaning, missing data imputation.

5.  Data Analysis – Descriptive, inferential, modeling.

6.  Interpretation – Contextualize findings (biological plausibility).

7.  Publication – Follow reporting guidelines (CONSORT, STROBE, PRISMA).

Advanced Note:

  • Reproducibility is key: Transparent reporting + open datasets + code sharing.

11. Example: Pediatric Pneumonia Trial

  • Research question: Is Penicillin alone (A) better than Penicillin + Chloramphenicol (B)?
  • Requires:
    • Sample size calculation (power analysis).
    • Randomization.
    • Blinding (to avoid bias).
    • Statistical comparison of groups (e.g., relative risk, confidence intervals).
    • Error analysis (misclassification of severity, measurement error in radiology).

 

12. Emerging Frontiers in Biostatistics

 

  • Bayesian statistics: Adaptive trials, hierarchical models.
  • Machine learning & AI: Predictive modeling, precision medicine.
  • Causal inference: Counterfactual reasoning, instrumental variables.
  • Big data epidemiology: Integrating EHR, genomic, and environmental data.
  • Systems medicine: Modeling complex biological networks.
Monday, 17 November 2025 18:07

AMPLE SIZE CONSIDERATIONS IN DATA ANALYSIS

Written by

 

 

1. QUICK RECAP: BASIC FORMULA (FOR PROPORTION)

For prevalence/proportion studies (very common in orthopaedics):

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Where:

  • n = sample size

  • Z = Z-score for confidence level (usually 1.96 for 95%)

  • p = expected prevalence (proportion)

  • q = 1 − p

  • d = margin of error (precision)

If p is unknown → use 0.5 (50%)
If population 10,000 → apply finite population correction (FPC)
If non-response expected → divide by (1 − non-response rate)

2. CASE 1: POPULATION > 10,000, UNKNOWN PREVALENCE

Case Study 1

Research question:
“What is the prevalence of chronic low back pain among adults attending the orthopaedic clinic at a county referral hospital?”

  • The hospital sees about 30,000 adult patients per year → treat as >10,000 (large population).

  • You don’t know the prevalence of chronic low back pain → use p = 0.5 (50%) to be safe.

Given:

  • Confidence level = 95% → Z = 1.96

  • p = 0.5

  • q = 1 − 0.5 = 0.5

  • d = 0.05

Step-by-step calculation (more than 3 steps)

Step 1: Write the formula

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Step 2: Substitute the values

n=(1.96)2⋅0.5⋅0.5(0.05)2n = \frac{(1.96)^2 \cdot 0.5 \cdot 0.5}{(0.05)^2}n=(0.05)2(1.96)2⋅0.5⋅0.5​

Step 3: Calculate Z²

1.962=3.84161.96^2 = 3.84161.962=3.8416

Step 4: Multiply p × q

0.5×0.5=0.250.5 \times 0.5 = 0.250.5×0.5=0.25

Step 5: Multiply the numerator

3.8416×0.25=0.96043.8416 \times 0.25 = 0.96043.8416×0.25=0.9604

Step 6: Calculate d²

(0.05)2=0.0025(0.05)^2 = 0.0025(0.05)2=0.0025

Step 7: Divide numerator by denominator

n=0.96040.0025=384.16n = \frac{0.9604}{0.0025} = 384.16n=0.00250.9604​=384.16

Step 8: Round up

n≈385 participantsn \approx 385 \text{ participants}n≈385 participants

Because the population is large (>10,000), NO finite correction is needed.
Next we will later see how to adjust this for non-response.

3. CASE 2: POPULATION > 10,000, KNOWN PREVALENCE

Case Study 2

Research question:
“What is the prevalence of post-operative wound infection following ORIF (open reduction internal fixation) at a regional orthopaedic centre?”

Suppose previous records show wound infection rate = 10% (p = 0.10) among ORIF patients.

Given:

  • Population (yearly ORIF patients) ≈ 12,000 → >10,000

  • p = 0.10

  • q = 1 − 0.10 = 0.90

  • d = 0.03 (you want ±3% precision)

  • Z = 1.96

Step-by-step calculation

Step 1: Formula

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Step 2: Substitute values

n=(1.96)2⋅0.10⋅0.90(0.03)2n = \frac{(1.96)^2 \cdot 0.10 \cdot 0.90}{(0.03)^2}n=(0.03)2(1.96)2⋅0.10⋅0.90​

Step 3: Compute Z²

1.962=3.84161.96^2 = 3.84161.962=3.8416

Step 4: Compute p × q

0.10×0.90=0.090.10 \times 0.90 = 0.090.10×0.90=0.09

Step 5: Multiply for numerator

3.8416×0.09=0.3457443.8416 \times 0.09 = 0.3457443.8416×0.09=0.345744

Step 6: Compute d²

0.032=0.00090.03^2 = 0.00090.032=0.0009

Step 7: Divide numerator by denominator

n=0.3457440.0009=384.16n = \frac{0.345744}{0.0009} = 384.16n=0.00090.345744​=384.16

Step 8: Round up

n≈385 ORIF patientsn \approx 385 \text{ ORIF patients}n≈385 ORIF patients

Again, no finite population correction because population is large (> 10,000).

4. CASE 3: POPULATION 10,000, KNOWN PREVALENCE (USE FPC)

Now we bring in the finite population correction (FPC).

Case Study 3

Research question:
“What is the prevalence of knee osteoarthritis among patients aged ≥50 years attending your rural orthopaedic clinic in one year?”

  • Total number of patients aged ≥50 years in that year = N = 2,400

  • Previous small audit suggests OA prevalence = 40% (p = 0.40)

Given:

  • N = 2,400 (10,000 → we MUST correct)

  • p = 0.40

  • q = 0.60

  • d = 0.05

  • Z = 1.96

Step-by-step calculation – Part 1: Initial n (as if population is large)

Step 1: Formula for n

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Step 2: Substitute

n=(1.96)2⋅0.40⋅0.60(0.05)2n = \frac{(1.96)^2 \cdot 0.40 \cdot 0.60}{(0.05)^2}n=(0.05)2(1.96)2⋅0.40⋅0.60​

Step 3: Compute Z²

1.962=3.84161.96^2 = 3.84161.962=3.8416

Step 4: Compute p × q

0.40×0.60=0.240.40 \times 0.60 = 0.240.40×0.60=0.24

Step 5: Multiply numerator

3.8416×0.24=0.9219843.8416 \times 0.24 = 0.9219843.8416×0.24=0.921984

Step 6: Compute d²

0.052=0.00250.05^2 = 0.00250.052=0.0025

Step 7: Calculate n

n=0.9219840.0025=368.79n = \frac{0.921984}{0.0025} = 368.79n=0.00250.921984​=368.79

Round → n ≈ 369 (initial)

Step-by-step calculation – Part 2: Apply Finite Population Correction (FPC)

nadj=n1+n−1Nn_{adj} = \frac{n}{1 + \frac{n - 1}{N}}nadj​=1+Nn−1​n​

Step 1: Substitute n and N

nadj=3691+369−12400n_{adj} = \frac{369}{1 + \frac{369 - 1}{2400}}nadj​=1+2400369−1​369​

Step 2: Simplify numerator in the fraction

369−1=368369 - 1 = 368369−1=368 3682400=0.15333‾≈0.1533\frac{368}{2400} = 0.1533\overline{3} \approx 0.15332400368​=0.15333≈0.1533

Step 3: Add 1 in the denominator

1+0.1533=1.15331 + 0.1533 = 1.15331+0.1533=1.1533

Step 4: Divide n by this value

nadj=3691.1533≈319.9n_{adj} = \frac{369}{1.1533} \approx 319.9nadj​=1.1533369​≈319.9

Step 5: Round up

Final corrected sample size:

nadj≈320 patientsn_{adj} \approx 320 \text{ patients}nadj​≈320 patients

So, because the population is only 2,400, you don’t need 369; 320 is enough.

5. CASE 4: POPULATION 10,000, UNKNOWN PREVALENCE

Case Study 4

Research question:
“What is the prevalence of chronic non-union in long bone fractures seen over 1 year in your district hospital?”

You don’t know the prevalence.

  • Total fracture patients in 1 year: N = 1,000

  • Unknown prevalence → use p = 0.5

  • Z = 1.96

  • d = 0.05

Part 1: Calculate initial n assuming large population

Step 1: Formula

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Step 2: Substitute values

n=(1.96)2⋅0.5⋅0.5(0.05)2n = \frac{(1.96)^2 \cdot 0.5 \cdot 0.5}{(0.05)^2}n=(0.05)2(1.96)2⋅0.5⋅0.5​

Step 3: Compute Z²

1.962=3.84161.96^2 = 3.84161.962=3.8416

Step 4: Compute p × q

0.5×0.5=0.250.5 \times 0.5 = 0.250.5×0.5=0.25

Step 5: Numerator

3.8416×0.25=0.96043.8416 \times 0.25 = 0.96043.8416×0.25=0.9604

Step 6: d²

0.052=0.00250.05^2 = 0.00250.052=0.0025

Step 7: n

n=0.96040.0025=384.16n = \frac{0.9604}{0.0025} = 384.16n=0.00250.9604​=384.16

So initial n ≈ 385

Part 2: Apply FPC (N = 1,000)

nadj=n1+n−1Nn_{adj} = \frac{n}{1 + \frac{n - 1}{N}}nadj​=1+Nn−1​n​

Step 1: Substitute

nadj=3851+385−11000n_{adj} = \frac{385}{1 + \frac{385 - 1}{1000}}nadj​=1+1000385−1​385​

Step 2: Simplify

385−1=384385 - 1 = 384385−1=384 3841000=0.384\frac{384}{1000} = 0.3841000384​=0.384

Step 3: Add 1

1+0.384=1.3841 + 0.384 = 1.3841+0.384=1.384

Step 4: Divide

nadj=3851.384≈278.1n_{adj} = \frac{385}{1.384} \approx 278.1nadj​=1.384385​≈278.1

Step 5: Round up

nadj≈279 patientsn_{adj} \approx 279 \text{ patients}nadj​≈279 patients

So you only need 279 out of 1,000 fracture patients to estimate non-union prevalence.

6. CASE 5: UNKNOWN TOTAL POPULATION (TREAT AS VERY LARGE)

Sometimes you don’t know the exact population size:

  • Example: You study community prevalence of back pain in a town, but you don’t have accurate census data for adults.

In such cases, we treat population as very large and do NOT apply FPC.

Case Study 5

“Prevalence of chronic low back pain among boda-boda riders in a large city.”

  • No accurate count of all riders → population treated as infinite/very large.

  • Suppose previous study estimates p = 0.25 (25%).

Given:

  • p = 0.25 → q = 0.75

  • d = 0.05

  • Z = 1.96

Step 1: Formula

n=Z2⋅p⋅qd2n = \frac{Z^2 \cdot p \cdot q}{d^2}n=d2Z2⋅p⋅q​

Step 2: Substitute

n=(1.96)2⋅0.25⋅0.75(0.05)2n = \frac{(1.96)^2 \cdot 0.25 \cdot 0.75}{(0.05)^2}n=(0.05)2(1.96)2⋅0.25⋅0.75​

Step 3: Z²

1.962=3.84161.96^2 = 3.84161.962=3.8416

Step 4: p × q

0.25×0.75=0.18750.25 \times 0.75 = 0.18750.25×0.75=0.1875

Step 5: Numerator

3.8416×0.1875=0.72033.8416 \times 0.1875 = 0.72033.8416×0.1875=0.7203

Step 6: d²

0.052=0.00250.05^2 = 0.00250.052=0.0025

Step 7: n

n=0.72030.0025=288.12n = \frac{0.7203}{0.0025} = 288.12n=0.00250.7203​=288.12

Step 8: Round up → 289 riders

No FPC because population is treated as very large/unknown.

7. ADJUSTMENT FOR NON-RESPONSE (ALL CASES)

In real orthopaedic research, some patients will:

  • refuse consent

  • fail to return for follow-up

  • not complete questionnaires

We adjust using:

nfinal=nadj1−non-response raten_{final} = \frac{n_{adj}}{1 - \text{non-response rate}}nfinal​=1−non-response ratenadj​​

If non-response rate is 10% → use 0.90 in denominator.
If 20% → 0.80, etc.

Example Using Case Study 3 (Knee OA, N = 2400, nₐ𝚍ⱼ = 320)

Assume 10% non-response.

Step 1: Write adjustment formula

nfinal=nadj1−NRn_{final} = \frac{n_{adj}}{1 - NR}nfinal​=1−NRnadj​​

Step 2: Substitute

nfinal=3201−0.10n_{final} = \frac{320}{1 - 0.10}nfinal​=1−0.10320​

Step 3: Calculate denominator

1−0.10=0.901 - 0.10 = 0.901−0.10=0.90

Step 4: Divide

nfinal=3200.90=355.56n_{final} = \frac{320}{0.90} = 355.56nfinal​=0.90320​=355.56

Step 5: Round up

nfinal≈356 patientsn_{final} \approx 356 \text{ patients}nfinal​≈356 patients

So you plan to recruit 356 patients, expecting that about 10% may not respond or complete the study.

8. SHORT NARRATIVE SUMMARY FOR STUDENTS

 

  1. If population is very large (>10,000) or unknown
    → No finite correction
    → Use basic formula for n.

  2. If population is small (10,000)
    → First compute n (as if large).
    → Then apply FPC formula to reduce n.

  3. If prevalence is known from previous data
    → Use that p (e.g., 0.10, 0.25, 0.40).

  4. If prevalence is unknown
    → Use p = 0.5 (gives maximum, safest n).

  5. Always adjust for non-response
    → Divide by (1 − non-response rate).

  6. Always round sample size UP, not down.

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Saturday, 01 November 2025 19:06

SAMPLE SIZE DETERMINATION

Written by

 

SAMPLE SIZE DETERMINATION

🔹 Definition

Sample size determination is the process of calculating the minimum number of subjects or observations required in a study to:

  • Detect a true effect or difference,

  • Achieve statistical significance, and

  • Ensure adequate power and precision of results.

It prevents both underpowered studies (too few participants → false negatives) and wasteful studies (too many participants → unnecessary cost/time).

⚖️ Key Factors Affecting Sample Size

Factor Meaning Effect
Confidence Level (Z) Certainty we want (usually 95%) Higher confidence → Larger sample
Margin of Error (E) Acceptable level of error Smaller error → Larger sample
Population Variability (SD or p) How much the data vary Higher variability → Larger sample
Power (1−β) Ability to detect a true effect (usually 80% or 90%) Higher power → Larger sample
Expected Effect Size (Δ) Minimum difference considered clinically important Smaller difference → Larger sample

🧩 1. For Estimating a Mean

n=(Z×SDE)2n = \left( \frac{Z \times SD}{E} \right)^2n=(EZ×SD​)2

Where:

  • n = required sample size

  • Z = Z-score for desired confidence (1.96 for 95%)

  • SD = estimated standard deviation

  • E = margin of error (precision)

Example 1 (Mean)

A study aims to estimate mean fasting blood sugar.

  • SD = 15 mg/dL

  • Desired precision (E) = ±5 mg/dL

  • Confidence = 95% (Z = 1.96)

n=(1.96×155)2=(5.88)2=34.6n = \left( \frac{1.96 \times 15}{5} \right)^2 = (5.88)^2 = 34.6n=(51.96×15​)2=(5.88)2=34.6

✅ Minimum sample size = 35 participants

🧩 2. For Estimating a Proportion

n=Z2×p(1−p)E2n = \frac{Z^2 \times p(1-p)}{E^2}n=E2Z2×p(1−p)​

Where:

  • p = estimated proportion (from pilot or literature)

  • E = acceptable margin of error

  • Z = Z-score for desired confidence level

Example 2 (Proportion)

Prevalence of hypertension estimated at 30% (p = 0.3).
Margin of error = 5% (E = 0.05).
Confidence = 95% (Z = 1.96).

n=(1.96)2×0.3(1−0.3)(0.05)2=3.8416×0.210.0025=323.1n = \frac{(1.96)^2 \times 0.3(1 - 0.3)}{(0.05)^2} = \frac{3.8416 \times 0.21}{0.0025} = 323.1n=(0.05)2(1.96)2×0.3(1−0.3)​=0.00253.8416×0.21​=323.1

✅ Minimum sample size = 324 participants

🧩 3. For Comparing Two Groups (Means)

n=2×(Zα/2+ZβΔ/SD)2n = 2 \times \left( \frac{Z_{\alpha/2} + Z_{\beta}}{\Delta/SD} \right)^2n=2×(Δ/SDZα/2​+Zβ​​)2

Where:

  • Δ = expected difference between group means

  • Zα/2 = 1.96 for 95% confidence

  • Zβ = 0.84 for 80% power

Example 3 (Two Means)

Comparing two antihypertensive drugs:

  • Expected difference = 10 mmHg

  • SD = 15 mmHg

  • α = 0.05, β = 0.20 (power 80%)

n=2×(1.96+0.8410/15)2=2×(2.8×1.5)2=2×(4.2)2=2×17.64=35.28n = 2 \times \left( \frac{1.96 + 0.84}{10/15} \right)^2 = 2 \times (2.8 \times 1.5)^2 = 2 \times (4.2)^2 = 2 \times 17.64 = 35.28n=2×(10/151.96+0.84​)2=2×(2.8×1.5)2=2×(4.2)2=2×17.64=35.28

✅ n ≈ 36 per group

📊 4. Adjustments

  • For Non-response or Attrition:
    nadjusted=n(1−expected dropout rate)n_{adjusted} = \frac{n}{(1 - \text{expected dropout rate})}nadjusted​=(1−expected dropout rate)n​
    e.g., If dropout = 10%, then multiply by 1.11.

  • For Finite Population (N 10,000):
    nadj=n1+n−1Nn_{adj} = \frac{n}{1 + \frac{n - 1}{N}}nadj​=1+Nn−1​n​

💡 Practical Notes for Medical Research

 

  • Use pilot studies or previous literature to estimate SD or p.

  • For unknown proportions, assume p = 0.5 (gives largest sample).

  • Always justify sample size in proposals (Ethics & IRB requirement).

  • Many use software like OpenEpi, Raosoft, or Epi Info for precise calculations.

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Saturday, 01 November 2025 18:58

INTERVALS DERTEMINATION

Written by

 

🧮 Determination of Confidence Intervals

A Confidence Interval (CI) gives a range of values within which the true population parameter (mean, proportion, risk ratio, etc.) is expected to fall — with a specified level of confidence (e.g., 95%).

🔹 Step-by-Step Process

Step 1: Collect Data

Obtain sample data — for example:

  • Mean or proportion

  • Standard deviation (SD)

  • Sample size (n)

Step 2: Choose the Confidence Level

Decide how confident you want to be:

  • 90% → Z = 1.645

  • 95% → Z = 1.96

  • 99% → Z = 2.576

Higher confidence = wider interval (more certain, less precise).

Step 3: Calculate the Standard Error (SE)

For Mean:

SE=SDnSE = \frac{SD}{\sqrt{n}}SE=n​SD​

For Proportion:

SE=p(1−p)nSE = \sqrt{\frac{p(1-p)}{n}}SE=np(1−p)​​

Where:

  • p = sample proportion (e.g., 0.45 for 45%)

Step 4: Determine the Margin of Error (ME)

ME=Z×SEME = Z \times SEME=Z×SE

Step 5: Construct the Confidence Interval

Confidence Interval=Xˉ±ME\text{Confidence Interval} = \bar{X} \pm MEConfidence Interval=Xˉ±ME

or for proportions:

CI=p±Z×SECI = p \pm Z \times SECI=p±Z×SE

🔹 Example 1: Mean

A study on systolic BP (mmHg):

  • Mean (X̄) = 130

  • SD = 10

  • n = 25

  • Confidence Level = 95%

Step 1: SE = 10 / √25 = 10 / 5 = 2
Step 2: ME = 1.96 × 2 = 3.92
Step 3: CI = 130 ± 3.92 → (126.08 – 133.92 mmHg)

✅ Interpretation: We are 95% confident the true population mean lies between 126.08 and 133.92 mmHg.

🔹 Example 2: Proportion

In a vaccine study:

  • p = 0.80 (80% effectiveness)

  • n = 100

  • Confidence Level = 95%

Step 1: SE = √[0.8(1−0.8)/100] = √(0.16/100) = 0.04
Step 2: ME = 1.96 × 0.04 = 0.078
Step 3: CI = 0.80 ± 0.078 = (0.722 – 0.878) or 72.2%–87.8%

🔹 Key Insights

Factor Effect on CI
Larger sample (↑n) CI becomes narrower
Higher variability (↑SD) CI becomes wider
Higher confidence (e.g., 99%) CI becomes wider

🔹 Clinical Relevance

Confidence intervals are used to:

 

  • Evaluate precision of study results

  • Check statistical significance (if CI excludes 0 or 1)

  • Compare treatment effects between groups

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Saturday, 01 November 2025 18:21

Confidence Levels

Written by

 

Confidence Levels (in Medical Research)

Definition

A confidence level represents the degree of certainty that a population parameter (like a mean or proportion) lies within a specified confidence interval (CI) based on sample data.

It tells us how confident we are that our estimate from the sample reflects the true value in the population.

Common Confidence Levels

Confidence Level Confidence Interval Range Interpretation
90% Narrower CI 90% confident the true value lies within this range; allows more risk (10%) of error.
95% Standard in medical research 95% confident the true population value is within the CI; 5% risk of error.
99% Wider CI 99% confident; more conservative but less precise.

Example

If a study finds that a new antihypertensive drug reduces systolic BP by 10 mmHg (95% CI: 8–12 mmHg):

  • We are 95% confident the true average reduction lies between 8 and 12 mmHg.

  • There is a 5% chance that the true value lies outside this range.

Key Concepts

  • Confidence Level → Degree of certainty (e.g., 95%).

  • Confidence Interval → Range of values that likely contain the true value.

  • Higher Confidence Level → Greater certainty, but wider (less precise) interval.

  • Lower Confidence Level → Narrower interval, but less certainty.

Clinical Relevance

 

  • Helps in interpreting study reliability.

  • Used in clinical trials, diagnostic test evaluation, and epidemiological studies.

  • A 95% CI that does not include zero (or 1 for relative risk/odds ratio) suggests statistical significance.

  • 1. Relationship Between Confidence Level and Confidence Interval

    A confidence level (e.g., 95%) corresponds to how certain we are that the true population value lies within the confidence interval (CI) computed from the sample data.
    It is based on the sampling distribution of the mean (or proportion) and uses the standard error to estimate variability.

    🔹 2. General Formula for Confidence Interval (for a Mean)

    Confidence Interval=Xˉ±Z×SDn\text{Confidence Interval} = \bar{X} \pm Z \times \frac{SD}{\sqrt{n}}Confidence Interval=Xˉ±Z×n​SD​

    Where:

    • Xˉ\bar{X}Xˉ = sample mean

    • SD = standard deviation of the sample

    • n = sample size

    • Z = Z-score corresponding to the chosen confidence level

    🔹 3. Z-Scores for Common Confidence Levels

    Confidence Level Z-Score (from Normal Distribution) Two-Tail α (error)
    90% 1.645 0.10
    95% 1.96 0.05
    99% 2.576 0.01

    So, for a 95% confidence level, you use Z = 1.96.

    🔹 4. Example Calculation

    Example:

    A study measures systolic blood pressure reduction after a drug trial.

    • Mean reduction (Xˉ\bar{X}Xˉ) = 10 mmHg

    • SD = 4 mmHg

    • Sample size (n) = 25

    • Confidence level = 95%

    Step 1: Compute Standard Error (SE)

    SE=SDn=425=45=0.8SE = \frac{SD}{\sqrt{n}} = \frac{4}{\sqrt{25}} = \frac{4}{5} = 0.8SE=n​SD​=25​4​=54​=0.8

    Step 2: Compute Margin of Error (ME)

    ME=Z×SE=1.96×0.8=1.568ME = Z \times SE = 1.96 \times 0.8 = 1.568ME=Z×SE=1.96×0.8=1.568

    Step 3: Construct Confidence Interval

    CI=10±1.57=(8.43, 11.57)CI = 10 \pm 1.57 = (8.43,\ 11.57)CI=10±1.57=(8.43, 11.57)

    ✅ Interpretation: We are 95% confident that the true mean BP reduction lies between 8.43 mmHg and 11.57 mmHg.

    🔹 5. For Proportions

    When dealing with proportions (like prevalence rates or vaccine efficacy):

    CI=p±Z×p(1−p)nCI = p \pm Z \times \sqrt{\frac{p(1 - p)}{n}}CI=p±Z×np(1−p)​​

    Where p = sample proportion (e.g., 0.40 for 40%).

    🔹 6. Relation Between Confidence Level and Width

    • Increasing confidence (e.g., 95% → 99%) increases Z, thus widening the CI.

    • Larger samples (↑n) reduce SE, thus narrowing the CI.

    🔹 7. Conceptual Analogy

    If you repeated the study 100 times,

    • At a 95% confidence level, about 95 of the 100 intervals would contain the true population mean.

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Tuesday, 28 October 2025 09:54

gggg

Written by

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Tuesday, 28 October 2025 09:49

jjjj

Written by

 

 

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Tuesday, 28 October 2025 09:41

hhh

Written by

 

 

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Tuesday, 28 October 2025 04:11

normal distribution curve

Written by

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Monday, 27 October 2025 07:07

MEASURE OF DISPERSION

Written by

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...
Sunday, 26 October 2025 20:30

MEASURE OF CENTRAL TENDENCY

Written by

 

Published in BIOSTATISTICS FOR UPGRADING CLASS
Read more...

Overview

Health Africa is a tight-knit community of scientific, medical, and life sciences experts that produce and share the latest information, in a readable, understandable way. Also join our social nework group  Health Africa Conference for convenient communication. 

Information

Knowledge

Get Support

For All Enquiries Contact Us

  • Mombasa, Kenya
  • +254 713 521 930
  • info@healthafricalibrary.co.ke

Login Form

Register to save more

Register an account today and get great deals, offers and customised products only for you.

  • Secure registration form
  • Your data on your hand, you can update anytime
  • Save your favourite products and access anytime
  • Order products easily.
Create Account