Quantitative & Experimental Methods in Psychiatry – Comprehensive Study Notes

Page 1

• Topic Introduction
  • Course/Chapter: 5 Quantitative and Experimental Methods in Psychiatry, Section 5.1 (Epidemiology)
  • Authors: Diana E. Clarke et al.
  • Epidemiology = study of how often disease occurs in populations, temporal changes, determinants.
  • Psychiatric epidemiology aligns with chronic‐disease models (vs. acute/infectious).
  • Stuart L. Morris’s 1957 “seven uses of epidemiology” provide framework (historical study, community diagnosis, health-service utilization, completing the clinical picture, identification of syndromes, assessing individual risks, identifying causes).

Page 2

• Historical Study
  • Harder in psychiatry because of diagnostic changes (e.g., post-1970 DSM shifts); example: projected rise in Alzheimer disease due to ageing baby boomers.
• Community Diagnosis
  • Describes health status of population and burden (reduced productivity, cost, premature mortality).
  • WHO Global Burden of Disease (GBD) project introduced.
• Health-Service Utilization
  • Documents “de facto mental-health system” incl. primary care, social services, schools, alternative medicine.
  • Primary care is key worldwide—necessitates psychiatric knowledge among PCPs.

Page 3

• Treatment Need Conceptual Problems
  • Unlike fractures/sepsis, psychiatric disorders sit on continuum; diagnostic thresholds fuzzy.
  • Subthreshold symptoms may still cause disability; spontaneous remission possible.
  • Hence “need for treatment” remains unresolved yet unmet need remains large.
• Completing the Clinical Picture
  • Community studies capture natural history, prodrome, progression.
  • Clinical samples biased (Berkson bias) because treated cases tend to be more severe/comorbid.

Page 4

• Identification of Syndromes
  • Pre-DSM-III categories poorly defined; categorical symptom-based system has limits.
  • DSM-5 added cross-cutting dimensional assessments.
  • NIMH Research Domain Criteria (RDoC) explores neuroscience-based dimensions.
  • Epidemiology clarifies prevalence + disability; high prevalence/low disability likely normal variants.
  • Example: high comorbidity among depression/anxiety/somatization.

Page 5

• Assessing Individual Risks
  • Classic CVD analogy: cholesterol, hypertension, smoking.
  • Translation to individual psychiatric prevention still nascent though early treatment vital.
• Identifying Causes
  • Psychiatric disorders multifactorial; gene-environment interplay (e.g., 5-HTTLPR × stress → depression).
• Measures of Frequency: Prevalence Basics
  • Prevalence ratio needs: case definition, population, time.

Page 6

• Prevalence Types
  • Point / 1-Month (used more in psychiatry).
  • 1-Year (useful for service planning).
  • Lifetime (for risk-factor research but recall bias).
  • Demographic shifts require standardization when comparing rates.

Page 7

• Incidence Concepts
  • New events in a time frame.
  • Numerator: new cases; denominator excludes existing cases.
  • “First incidence” = first-ever cases.
  • Person-time denominators adjust for dynamic populations.
• Relationship: $P \approx I \times d$
  • Prevalence reflects incidence + duration; therapies shortening duration lower prevalence; life-prolonging treatments can raise prevalence (e.g., AIDS post-HAART).
• Comorbidity
  • Random vs. non-random co-occurrence; diagnostic artifacts (e.g., substance use in DSM-III ASPD).

Page 8

• Effect & Association Metrics
  • Risk factor ↑ likelihood; protective ↓.
  • Web of causation; rare outcomes need large cohorts.
  • 2×2 table cells a,b,c,d.
  • Relative Risk (RR): $RR = \dfrac{a/(a+b)}{c/(c+d)}$

Page 9

• Odds Ratio (OR)
  • Used in case–control; cross-product $OR = \dfrac{ad}{bc}$.
  • Interpretation around 1.
• Confounding
  • Factor associated with both exposure & outcome; adjust via stratification/regression.
• Mediation vs. Moderation (Effect Modification)
  • Mediation: A explains path E→O.
  • Moderation: effect of E on O differs by levels of A; tested via interaction.
• Attributable Fraction (AF)
  • Among exposed: $AF_e = \frac{RR-1}{RR}$
  • Population AF depends on exposure prevalence.

Page 10

• Study Designs Overview
  • Experimental vs. Observational.
• Experimental (Clinical Trials)
  • Efficacy trials (highly selected); Effectiveness trials (real-world).
  • Controlled, randomized, double-blind; ethical issues with placebos.
  • Prevention trials deal with incident cases; randomization sometimes infeasible (universal interventions).

Page 11

• Observational Analytic Studies
  • Cohort: select by exposure, follow for outcome.
  • Case-control: select by outcome, measure past exposure; OR approximates RR.
• Descriptive Studies
  • Cross-sectional, repeated cross-sectional, longitudinal; provide foundational prevalence/incidence data.

Page 12

• Sampling Fundamentals
  • Sampling frame; representative sample.
  • Probability vs. Non-probability sampling.
  • Simple random; Stratified (↑precision).
  • Non-probability: convenience, consecutive, quota, snowball.

Page 13

• Measurement
  • Instruments must match target population; caution with modifications.
  • Reliability
  • Test–retest; inter-rater (Cohen’s $\kappa$; ≥0.4 acceptable).
  • Validity
  • Content, Criterion (concurrent/predictive; sensitivity, specificity, PPV/NPV), Construct (Robins & Guze validators).

Page 14

• Case Identification & DSM Evolution
  • Self-report + informants; no pathognomonic tests.
  • DSM-I (1952) & DSM-II unreliable; U.S.–U.K. study showed diagnostic variance.
  • DSM-III (1980) introduced operational criteria; subsequent revisions (DSM-III-R, DSM-IV, DSM-5, DSM-5-TR).

Page 15

• ICD Relations & Harmonization
  • DSM-ICD linkage; ICD-11 (2019) aligned ~30% identical disorders with DSM-5.
• Disability Classification
  • DSM definition includes distress/impairment.
  • WHO ICF (2001): Functioning & Disability framework (Body Functions, Activities/Participation, Environmental factors).

Page 16

• Diagnostic Instruments – Adults
  • DIS (1978) for DSM-III; fully structured; paved way for ECA.
  • CIDI (WHO): global, computerized; UM-CIDI (NCS) modifications; WMH-CIDI (adds functioning, new disorders).
  • AUDADIS-5 (DSM-5) for substance + comorbid disorders.
  • SCID-5 (semi-structured; clinician-administered).

Page 17

• Diagnostic Instruments – Children/Adolescents
  • DISC (lay interview; parent/child versions; DSM-5 update DISC-5).
  • CAPA & variants (interviewer-based judgment).
  • K-SADS-PL DSM-5.
• Disability Scales
  • WHODAS 2.0 (36-item; 12-item short form).
  • CGAS / BIS for youth.

Page 18

• Major Epidemiologic Surveys – Adults
  • ECA (1980s; 5 sites; DSM-III DIS) ➔ comorbidity, unmet need.
  • NCS (1990-92; UM-CIDI) & NCS-R (2001-03); higher prevalence partly due to methods.
  • NSDUH (annual household; DSM-IV/5; 2019 data: 50k interviews).
  • WMH Initiative (29 countries; WMH-CIDI) ➔ cross-national prevalence range.
  • NESARC (NLAES, NESARC-I/II/III; AUDADIS).

Page 19

• Surveys – Youth
  • MECA study (methods, DISC reliability).
  • NCS-A (2001-04; 13-17 yrs): 1-year any disorder 40.2%.
  • YRBSS (behaviors; 4.9 million students since 1991).
  • NSCH & other CDC surveys provide mental-health modules.

Page 20

• Global Burden of Disease (GBD)
  • DALY = $YLL + YLD$ (years life lost + years lived disabled).
  • Mental disorders ≈5% of total DALYs (depression > anxiety > SUD> schizophrenia).
  • Trends 1990-2019: rising DALYs for mental disorders except alcohol use.

Page 21

• Emerging Issue: COVID-19 Mental-Health Impact
  • ↑ prevalence of depression, anxiety, eating-disorder pathology, burnout, PTSD, violence.
  • Risk factors: frontline worker stress, sleep disturbance, loneliness.
  • Protective: structured routines, nature, healthy sleep, limited news exposure.

Page 22 (Section 5.2 Statistics & Experimental Design)**

• Why Psychiatrists Need Stats
  • Critical appraisal beyond abstracts; statistics = lingua franca.
• Basics Refresher
  • Probability, sample space, random variable, PDF/CDF, mean, median, variance.
  • Independence & conditional probability; $P(X|Y)=P(X)$ if independent.

Page 23

• Estimation
  • Maximum Likelihood Estimator (MLE).
  • Properties: unbiasedness, consistency, efficiency.
  • Confidence Interval (CI): $P[a\le\phi\le b]=c$ (frequentist interpretation).

Page 24

• Hypothesis Testing
  • Null vs. Alternative; type I ($\alpha$), type II ($\beta$), power $(1-\beta)$.
  • P value = probability (under null) of observing statistic ≥ extreme as seen.
  • Multiple testing ➔ Family-Wise Error Rate (FWER); Bonferroni, Sidak, Holm step-down, Hochberg step-up, closed testing, min-test.
  • False Discovery Rate (FDR) controls expected proportion false positives (e.g., Benjamini–Hochberg).

Page 25

• Bayesian Approach
  • Bayes theorem: $P(\phi|x)=\dfrac{f(x|\phi)P(\phi)}{\int f(x|\theta)P(\theta)d\theta}$.
  • Prior (subjective/objective), Posterior distribution; conjugate priors facilitate updating.

Page 26

• Causality & Randomization
  • Potential outcome framework (Rubin): each subject has $O<em>T, O</em>P$; causal effect $E(O<em>T-O</em>P)$.
  • Random assignment ensures $E(O<em>T|T)-E(O</em>P|P)=E(O<em>T-O</em>P)$.
• Matching & Propensity Scores
  • When randomization infeasible; estimate probability of treatment assignment, match / stratify on propensity.

Page 27

• Randomization Tests & Rank-based variants
  • Permutation of labels to derive exact sampling distribution; compute P value.
• Likely Responder Analysis
  • Identify responder subgroup first, then test; uses potential outcome & propensity methods.

Page 28

• Handling Missing Data
  • Mechanisms: MCAR, MAR, Non-ignorable.
  • Simple deletions vs. imputation (mean substitution, hot-deck, multiple imputation, EM algorithm, model-based).
  • Longitudinal → LOCF cautions; HLM / mixed models better.
  • Pattern-mixture models for non-ignorable missingness.

Page 29

• Regression & Classification
  • Linear regression assumptions; OLS vs. MLE.
  • Logistic regression: $\log\frac{p}{1-p}=\beta_0+\beta x$.
  • Machine Learning: CART, Random Forests, Support Vector Machines; ROC & AUC.

Page 30

• Contingency Tables & Log-Linear Models
  • Chi-square, CMH test, Breslow–Day for homogeneity of ORs.
  • GLM unifies regression for exponential family; link functions.

Page 31

• Longitudinal Models
  • Repeated measures ANOVA limitations.
  • Mixed/HLM models (random effects, flexible covariance).
  • Generalized Estimating Equations (GEE).
  • Latent Class Growth Analysis & Growth Mixture Modeling uncover trajectory subtypes.

Page 32

• Survival Analysis
  • Kaplan–Meier, hazards, log-rank test.
  • Cox proportional hazards model: $h(t|x)=h_0(t)\exp(\beta x)$.
  • Cure models for mixed susceptible/non-susceptible populations.

Page 33

• Study Design Considerations
  • Experimental design optimality; example: crossover trials, carry-over effect, AB/BA vs. extended designs.

Page 34

• Efficacy vs. Effectiveness vs. Cost-Effectiveness
  • Efficacy trials = controlled conditions; Effectiveness = real-world.
  • Cost-Effectiveness Analysis (CEA)
  • Program $P=(\varepsilon, \gamma)$.
  • CER $=\gamma/\varepsilon$; ICER $=(\Delta \gamma)/(\Delta \varepsilon)$.
  • Net Health Benefit $NHB<em>k(\lambda)=\varepsilon</em>k-\gamma_k/\lambda$.
  • Decision rules under fixed budget or WTP threshold $\lambda$.
  • Statistical CEA: bootstrapping, CIs for ICER, CMCB, Bayesian NHB.

Page 35

• Key Take-Home Messages
  • Understanding epidemiologic measures, statistical designs, and analytic methods is essential for modern psychiatric research.
  • Proper case definition, sampling, measurement reliability/validity, and handling of missing data safeguard inference.
  • Multiple sophisticated tools—Bayesian methods, machine learning, hierarchical models—expand capability but necessitate rigorous validation.