UW Slide Template Library

Reusable Patterns for Academic & Teaching Slides

Carlos Pineda-Antunez

How to use this template library

  • This file is not your talk, it’s a library of examples
  • Copy–paste any slide you like into your actual presentation .qmd
  • All slides inherit the UW styling from uw.css (purple, gold, bottom bar, etc.)
  • You can safely delete sections you don’t need

Learning objectives (example slide)

By the end of this section, participants should be able to:

  • Explain the intuition behind [method / concept]
  • Describe the key assumptions required for valid inference
  • Implement a simple example in R and interpret the output
  • Discuss when this method is appropriate in HEOR / health policy

Key definition / concept slide

Use this for central definitions like ICER, odds ratio, etc.

Definition: Incremental Cost-Effectiveness Ratio (ICER)

The ICER compares two strategies, 1 and 0:

[ = ]

where (C) is cost and (E) is effectiveness (e.g., QALYs).

  • If the denominator is small, ICERs can be unstable
  • We often complement ICERs with Net Monetary Benefit (NMB)
  • Choice of WTP threshold is crucial for interpretation

Worked example / derivation

You can reveal derivation steps gradually.

We want to estimate the effect of \(X\) on \(Y\):

\[ Y_i = \beta_0 + \beta_1 X_i + \varepsilon_i \]

  1. Assume \(\mathbb{E}[\varepsilon_i \mid X_i] = 0\)
  2. Use least squares to choose \(\hat{\beta}_0, \hat{\beta}_1\)
  3. The slope estimator can be written as: \[ \hat{\beta}_1 = \frac{\sum_i (X_i - \bar{X})(Y_i - \bar{Y})} {\sum_i (X_i - \bar{X})^2} \]
  4. Interpretation (HEOR):
    “A one-unit increase in \(X\) is associated with an average change of \(\hat{\beta}_1\) units in \(Y\), under the model assumptions.”

Quick quiz / check your understanding

Question

A logistic regression model estimates the log-odds of screening uptake.
Which of the following is not an appropriate interpretation?

  • A. The exponentiated coefficient gives an odds ratio
  • B. The model assumes a logit link between predictors and probabilities
  • C. The linear predictor can be interpreted as a risk difference
  • D. We can obtain predicted probabilities from the model

Answer:
The incorrect statement is C (the linear predictor is on the log-odds scale, not a risk difference).

Activity / discussion slide

Use this to pause and let students think or discuss.

  1. In your own work, what binary outcomes could be modeled with logistic regression?
  2. Which assumptions would you worry about (e.g., linearity in the logit, independence)?
  3. How would you communicate model results to a non-technical policymaker?

Be ready to share one concrete example from your discussion.

Code + output (R example)

Example: logistic regression for CRC screening uptake.

library(dplyr)

set.seed(123)
n <- 200

dat <- tibble(
  age      = rnorm(n, mean = 55, sd = 7),
  insured  = rbinom(n, 1, 0.7),
  screened = rbinom(n, 1, plogis(-5 + 0.06 * age + 0.8 * insured))
)

fit <- glm(screened ~ age + insured,
           data = dat,
           family = binomial)

summary(fit)

Call:
glm(formula = screened ~ age + insured, family = binomial, data = dat)

Coefficients:
            Estimate Std. Error z value Pr(>|z|)    
(Intercept) -6.27215    1.56494  -4.008 6.13e-05 ***
age          0.08266    0.02651   3.118  0.00182 ** 
insured      0.84025    0.40401   2.080  0.03755 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 229.22  on 199  degrees of freedom
Residual deviance: 214.30  on 197  degrees of freedom
AIC: 220.3

Number of Fisher Scoring iterations: 4
  • age has a positive association with screening probability
  • Being insured increases the odds of screening
  • Coefficients can be exponentiated to get odds ratios

Table slide with smaller font

# A tibble: 1 × 5
      n mean_age sd_age prop_ins prop_scr
  <int>    <dbl>  <dbl>    <dbl>    <dbl>
1   200     54.9   6.60    0.695     0.26

Tip: the class: small option shrinks the font size for dense tables.

Interactive tabset (multiple tables)

# A tibble: 1 × 5
      n mean_age sd_age prop_ins prop_scr
  <int>    <dbl>  <dbl>    <dbl>    <dbl>
1   200     54.9   6.60    0.695     0.26
# A tibble: 3 × 5
  term        estimate std.error statistic   p.value
  <chr>          <dbl>     <dbl>     <dbl>     <dbl>
1 (Intercept)  -6.27      1.56       -4.01 0.0000613
2 age           0.0827    0.0265      3.12 0.00182  
3 insured       0.840     0.404       2.08 0.0375

Interactive plot (plotly scatter)

You can replace this with any plotly object (e.g., CE planes, PSA results, etc.).

Embedding a Shiny app via iframe

References (manual list)

Use this version if you don’t want BibTeX / citation keys.

  • Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press; 2006.
  • Jansen JP, et al. Indirect treatment comparison / network meta-analysis… Value Health. 2014;17(2):157–173.
  • Hunink MGM, et al. Decision Making in Health and Medicine. Cambridge University Press; 2014.

References with citation keys (if using BibTeX)

class: uw-ref-slide

If you want Quarto to build references automatically:

  1. Create a refs.bib file in the same folder
  2. Add bibliography: refs.bib and link-citations: true to the YAML
  3. Use citation keys like this in your text: @briggs2006decision, @jansen2014indirect, etc.

Then add this slide at the end:

Styling (e.g., purple background) can be handled in uw.css for .uw-ref-slide.

Key takeaways slide

Use this to close a section or the whole talk.

  • Statistical models are tools to answer specific questions, not magic boxes
  • Always check whether the assumptions are at least somewhat plausible
  • Complement ICERs with NMB and transparent uncertainty communication
  • For teaching: mix theory, code, and interpretation with interactive elements
UW Presentation
Sections