Change to poisson dist to see if it is the glm fit to be unstable (#1059

)

This is not a fix. We need tomorrow's integration tests to see if also
the fit changes in the integration tests

tests/testthat/_snaps/summarize_glm_count.md

@@ -65,13 +65,79 @@
 
 # h_ppmeans works with healthy input
 
+    Code
+      fits
+    Output
+      $glm_fit
+      
+      Call:  stats::glm(formula = formula, family = stats::poisson(link = "log"), 
+          data = .df_row, offset = offset)
+      
+      Coefficients:
+               (Intercept)           REGION1Asia        REGION1Eurasia  
+                   2.01066               0.07631               0.64426  
+             REGION1Europe  REGION1North America  REGION1South America  
+                   2.13097              -0.07450               0.38102  
+                ARMCDARM B            ARMCDARM C  
+                   0.11048              -0.17694  
+      
+      Degrees of Freedom: 199 Total (i.e. Null);  192 Residual
+      Null Deviance:	    983.8 
+      Residual Deviance: 939 	AIC: 1498
+      
+      $emmeans_fit
+       ARMCD rate    SE  df asymp.LCL asymp.UCL
+       ARM A 12.6 1.238 Inf     10.43      15.3
+       ARM B 14.1 1.285 Inf     11.81      16.9
+       ARM C 10.6 0.971 Inf      8.85      12.7
+      
+      Results are averaged over the levels of: REGION1 
+      Confidence level used: 0.95 
+      Intervals are back-transformed from the log scale 
+      
+
+---
+
+    Code
+      fits2
+    Output
+      $glm_fit
+      
+      Call:  stats::glm(formula = formula, family = stats::quasipoisson(link = "log"), 
+          data = .df_row, offset = offset)
+      
+      Coefficients:
+               (Intercept)           REGION1Asia        REGION1Eurasia  
+                   2.01066               0.07631               0.64426  
+             REGION1Europe  REGION1North America  REGION1South America  
+                   2.13097              -0.07450               0.38102  
+                ARMCDARM B            ARMCDARM C  
+                   0.11048              -0.17694  
+      
+      Degrees of Freedom: 199 Total (i.e. Null);  192 Residual
+      Null Deviance:	    983.8 
+      Residual Deviance: 939 	AIC: NA
+      
+      $emmeans_fit
+       ARMCD rate   SE  df asymp.LCL asymp.UCL
+       ARM A 12.6 5.20 Inf      5.65      28.3
+       ARM B 14.1 5.39 Inf      6.68      29.8
+       ARM C 10.6 4.07 Inf      4.98      22.5
+      
+      Results are averaged over the levels of: REGION1 
+      Confidence level used: 0.95 
+      Intervals are back-transformed from the log scale 
+      
+
+---
+
     Code
       res
     Output
                      rate asymp.LCL asymp.UCL            ARM
-      A: Drug X      3.07  2.202774  4.278651      A: Drug X
-      B: Placebo     3.07  2.202774  4.278651     B: Placebo
-      C: Combination 3.07  2.202774  4.278651 C: Combination
+      A: Drug X      3.07  2.836527   3.32269      A: Drug X
+      B: Placebo     3.07  2.836527   3.32269     B: Placebo
+      C: Combination 3.07  2.836527   3.32269 C: Combination
 
 # s_glm_count works with healthy input
 
@@ -87,7 +153,7 @@
       [1] "Adjusted Rate"
       
       $rate_ci
-      [1] 1.983340 6.127155
+      [1] 3.047667 3.987387
       attr(,"label")
       [1] "95% CI"
       
@@ -121,7 +187,7 @@
       [1] "Adjusted Rate"
       
       $rate_ci
-      [1] 1.983340 6.127155
+      [1] 3.047667 3.987387
       attr(,"label")
       [1] "95% CI"
       
@@ -131,12 +197,12 @@
       [1] "Adjusted Rate Ratio"
       
       $rate_ratio_ci
-      [1] 0.3974979 0.3067371 2.0169939 1.8347687
+      [1] 0.7378778 0.6062152 1.0865633 0.9283695
       attr(,"label")
       [1] "95% CI"
       
       $pval
-      [1] 0.7897470 0.5287652
+      [1] 0.263119218 0.008203621
       attr(,"label")
       [1] "p-value"
       

tests/testthat/test-summarize_glm_count.R

@@ -169,11 +169,23 @@ testthat::test_that("h_ppmeans works with healthy input", {
   anl$AVAL_f <- as.factor(anl$AVAL)
 
   fits <- h_glm_count(
+    .var = "AVAL",
+    .df_row = anl,
+    variables = list(arm = "ARMCD", offset = "lgTMATRSK", covariates = c("REGION1")),
+    distribution = "poisson"
+  )
+
+  testthat::expect_snapshot(fits)
+
+  fits2 <- h_glm_count(
     .var = "AVAL",
     .df_row = anl,
     variables = list(arm = "ARMCD", offset = "lgTMATRSK", covariates = c("REGION1")),
     distribution = "quasipoisson"
   )
+
+  testthat::expect_snapshot(fits2)
+
   result <- h_ppmeans(
     obj = fits$glm_fit,
     .df_row = anl,
@@ -182,7 +194,7 @@ testthat::test_that("h_ppmeans works with healthy input", {
   )
 
   res <- testthat::expect_silent(result)
-  testthat::expect_snapshot(res)
+  testthat::expect_snapshot(res) # diff
 })
 
 testthat::test_that("s_glm_count works with healthy input", {
@@ -198,12 +210,12 @@ testthat::test_that("s_glm_count works with healthy input", {
     .in_ref_col = TRUE,
     variables = list(arm = "ARMCD", offset = "lgTMATRSK", covariates = c("REGION1")),
     conf_level = 0.95,
-    distribution = "quasipoisson",
+    distribution = "poisson",
     rate_mean_method = "ppmeans"
   )
 
   res <- testthat::expect_silent(result)
-  testthat::expect_snapshot(res)
+  testthat::expect_snapshot(res) # diff
 })
 
 testthat::test_that("s_glm_count works with no reference group selected.", {
@@ -221,12 +233,12 @@ testthat::test_that("s_glm_count works with no reference group selected.", {
       filter(ARMCD == "ARM B"),
     variables = list(arm = "ARMCD", offset = "lgTMATRSK", covariates = c("REGION1")),
     conf_level = 0.95,
-    distribution = "quasipoisson",
+    distribution = "poisson",
     rate_mean_method = "ppmeans"
   )
 
   res <- testthat::expect_silent(result)
-  testthat::expect_snapshot(res)
+  testthat::expect_snapshot(res) # diff
 })
 
 testthat::test_that("s_glm_count fails wrong inputs", {