Passionate about leveraging tech & AI for positive change, I specialize in developing innovative solutions to complex problems.

Smriti Mishra

For 14 years I've been developing my skills in clinical trials biostatistics (frequentist; non-Bayesian) across diverse therapeutic areas, supported by advanced statistical programming in R. I'm experienced in integrating R into controlled, numerically validated environments. ℹ️ I'm a self-taught applied biostatistician - I hold no degree in statistics or mathematics ⛔ No SAS - only R.  ⬇️ Details:

⧮ Experience (years):
⦿ Biostatistics: 14
⦿ Writing SAPs: 12
⦿ Programming in R: 22
⦿ TFL shells: 12
⦿ SDTM & ADaM: 2

⧮ Trials
⦿ RCT phase IIa/b (PK), III
⦿ RWE: pro-/retrospective observational

⧮ Therapeutic areas: asthma · oncology · rheumatology · osteoporosis · orthopaedics (osteogenesis, skeletal events) · dyslipidemia · anaemia · cardiology: rhythm disorders + cardiovascular diseases · thrombophilia · periodontology 

⧮ Scope of activity:
trial design: fixed & adaptive; multi-arm, cross-over/parallel + sample size & power · SAP+TFLs · randomization · analysis, programming, report · SOP

⧮ Selected topics in statistics:
⦿ Descriptive & inferential statistics: estimation & testing hypotheses (+ MCID: non-inferiority, equivalence, superiority).
⦿ Effect size: a variety of measures 
⦿ FWER control: CTP classic (MVT, Bonferroni, Holm, etc) and sequential (fixed/fallback/gatekeeping), graphical; FDR
⦿ Modern non-parametric, robust, bootstrap & permutation methods
⦿ Model-based hypothesis testing; LM, GLM (+GEE & GLMM), LQM[M], GAM
⦿ Longitudinal analysis ([c]LDA, ANCOVA post/change), pre-post: conditional (mixed models) & population-average (MMRM: GLS/GEE; GEE-GLM; Firth/Exact)
⦿ AME/MEM (LS/EM-means); G-calc
⦿ Contingency + questionnaires (+ Likert)
⦿ Type 2/3 analysis of main and interaction effects (n-way [RM]-AN[C]OVA). Simple effects / planned contrasts. · Wald, LRT, Rao
⦿ PCA, (E)FA
⦿ Missing data: RBMI/MICE/DR/*OCF, MNAR sensitivity / tipping-point
⦿ Estimand framework
⦿ Survival analysis:
* Scope:
- Estimation & comparison of: survival prob./CIF, quantiles, RMST · [weighted] LogRank/Fleming-Harrington: Max-Combo/Tarone-Ware/Gehan-Breslow/Peto; Gray
- Regression: (semi-)parametric · (non-)proportional hazards · time-dependent covariates · unequal effects & baseline hazards (strata) · recurrent events & competing risks (counting models, sub-distribution) + combined (joint frailty)
* Selected tools: Kaplan-Meier/Nelson-Aalen/Aalen-Johansen · Cox (+extensions: AG/PWP/FG), AFT, Frailty Models · landmark an. · MCF
⦿ SMR + inference
⦿ NTB via GPW; Buyse test
⦿ PharmacoKinetics
⦿ G-calc, IPTW, matching

⧮ Guidelines: ICH, FDA, RECIST, STARD, CONSORT, STROBE

Adrian Olszewski

Why is ANOVA called ANOVA (analysis of variance)? And what, the heck, is Analysis of Deviance?

➡️ For any kind of model (quantile, Cox, Poisson, gamma, Tobit, and any other kind of regression), to test the main effects we can apply Wald's approach to testing certain contrasts of the model coefficients per each factor or interaction. So we answer the question: "does the levels of factor X affect jointly the response of interest?" for ANY KIND OF DATA.

➡️ For likelihood- based models, like the Generalized Linear Model (GLM: (linear, Poisson, gamma, logistic, beta and more!), to test the main effects we can use Rao score or Likelihood Ratio assessment of the reduction of deviance per each factor or interaction. So it's the Analysis of Deviance - ADOVA

/ Wald's way holds too - always. For any kind of model. It only may be inefficient for smaller samples. /

➡️ For the General Linear Model (=GLM with Gaussian response and identity link, fit via OLS or GLS), deviance becomes the residual variance. So we assess the reduction in residual variance. So it's the Analysis of Variance - ANOVA

/ Here Wald = LRT = Rao ➡️ F statistic /

---
Yes, it's that easy!: you assess if adding a concrete factor reduced the residual (unexplained) variance enough. This "reduced enough" is evaluated with F statistic per factor.

For more complex but still likelihood-based models you need to generalize residual variance to deviance. It's reduction is tested with Chi2 statistic (Chi2 is limiting distribution for F under infinite denominator degrees of freedom).

For non-likelihood models, when you cannot use either, you can always test jointly appropriate combination of model coefficients - with Wald tests and Chi2 statistic. It's your life-saver.


#anova #statistics #datascience

Clinical Trials Biostatistician at 2KMM (100% R-based CRO) ⦿ Frequentist (non-Bayesian) paradigm ⦿ NOT a Data Scientist (no ML/Big data) ⦿ Against anti-car/-meat/-cash/-house/C40 restrictions

taplio

The Battle

Smriti Mishra

Adrian Olszewski

Followers

Posts

Engagements

Likes

Performance ScoreAdrian Olszewski's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Smriti Mishra

Adrian Olszewski

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Adrian Olszewski's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAdrian Olszewski's LinkedIn Performance