katieleap / normanal

Shiny app that calculates power for cluster-randomized trials, including running simulations and allowing for a coefficient of variation

normanal

normanal's Issues

Add a by-line, disclaimer, and (if it's not there yet) a license

By-line is definitely both of us!!

Changes for simulation tab

• Grey out calculate button
• Radio button for normal, poisson, dichotomous
• Change the inputs for each outcome distribution

Text should say number of clusters M in each arm

Alpha?

The table abbreviations shouldn't appear until there is a table.

Add a citations tab

This is what it should contain:

For Normal results with constant cluster size:

Donner, A. and Klar, N. Statistical Considerations in the Design and Analysis of Community Intervention Trials. The Journal of Clinical Epidemiology 1996;49(4):435-439 PMID: 8621994

Murray D. Design and Analysis of Group-Randomized Trials. New York: Oxford University Press; 1998

For variable cluster size using the coefficient of variation:

Eldridge SM, Ashby D, Kerry S. Sample size for cluster randomized trials: effect of coefficient of variation of cluster size and analysis method. Int. J. Epidemiol. (October 2006) 35(5): 1292-1300.doi: 10.1093/ije/dyl129 PMID: 16943232

For simulations:

Reich NG, Myers JA, Obeng D, Milstone AM, Perl TM. Empirical Power and Sample Size Calculations for Cluster-Randomized and Cluster-Randomized Crossover Studies. PLoS ONE 2012.7(4):e35564 doi: 10.1371/journal.pone.0035564 PMCID: PMC3338707

In addition, Katie, the Eldridge and Reich links could be hidden with, say, asterisks links, and revealed when clicked, like the model equation.

For the simulation, could we estimate the run time, and make sure the user really wants to do it?

And/or:

1. tell them no, if it'll take too long

or

2. spit out the R code for the user to input and run on their own R console?

Add a title in the UI, above the model equation

Normal outcome, two groups, equal number of clusters in each group.

Check to see whether the table values are rounding correctly

Also: make sure we have the alpha doing two-sided correctly

n.sim should be reactive

If the user is using ICC, the multiple values thingy should say "Use multple values for ICC".

And if they are using sig_b^2, it should say "Use multiple values for sig_b^2

Convergence failure issue

Create a method for when lmer returns a failure to converge error

when you uncheck the variable cluster size box reset the CV to b 0 and close the entry field

If you "clear all saved values", you can't calculate any more.

In fact, the whole thing is broken, maybe?

In "abbreviations", show calculation of ICC from sigb2 and sig2

Integration of CV and CVb

CV only works for ICC input currently.

Simulations are broken!!

Another next thing

Simulations for Poisson and logistic models

Table for the simulation results

Table is showing same values for approx and analytic power

Plot doesn't work with variable cluster size?

Can we include an "interrupt" button for the simulations?

Did we talk about this already?

Add SPHHS logo

Next up for this project: add option to the simulation

Allow a) specific values for the persons/cluster or b) distributions (NB, Poisson, integer uniform)

Fix the funny formatting in the citation tab

Plot problems

• Plot not dependent on calculate
• Sigma-b input doesn't have groupcheckbox selections fixed
• Switching should save values client side and clear them server side
• CV not disappearing after unchecked
• Don't plot approximate power (no legend) Values should be reactive to input, ICC/Sigma-b
• Larger points, fonts
• Fit a line? smoother function, deal with extreme values?
• Title: incorporate the input values
• Save as image

Translation between sigma_b^2 and ICC for logistic and Poisson