Running with and without a filter for Excl_criteria_ok results in differing values

filter(Excl_criteria_ok == 1) %>%  # TODO If no filter, there is an 'extra' one
  select(`UR number`, starts_with("Total_no_")) %>%
  mutate(
    Total_no_cr_epis = if_else(
      is.na(Total_no_cr_epis), " 0 cr epis", sprintf("%2d cr epis", Total_no_cr_epis)),
    Total_no_olig_epis = if_else(
      is.na(Total_no_olig_epis), " 0 olig epis", sprintf("%2d olig epis", Total_no_olig_epis)),
  ) %>%
  group_by(Total_no_cr_epis, Total_no_olig_epis) %>%
  summarise(
    Admissions = n(),
  ) %>%
  ungroup() %>%
  pivot_wider(names_from = Total_no_olig_epis, values_from = Admissions) %>%
  adorn_totals(c("row", "col")) %>% 

• Determine source of error
• Robustly correct error
• Update table to give Excl criteria not ok as a row as well

• Create a score based on multivariable model to predict probability of AKI (e.g. weight score)
• Clinically relevant explanation

True and false positive rates

Another approach to assessing value is to calculate the change in true and false positive rates at a clinical relevant risk threshold…which here could be 15% and/or 20%

• Calculate TP and FP rates for with/without cr_change
• Clinical relevance and risk threshold

AUC 95% CI

All AUCs have 95% Cis and we should show that the Cis of this model and of the one below do not overlap with the clinical model in isolation.

• Calculate 95% CIs for with/without cr_change
• Ensure they do not overlap

Net Reclassification Index

We should consider the net reclassification index (NRI)

• Calculate NRI

Brier Score

Consider calculating the Brier score. The Brier Score is a strictly proper score function or strictly proper scoring rule that measures the accuracy of probabilistic predictions.

• Calculate Brier score

• Figure 3: Change one of the blue or purple lines to red
• Figure 3: For the optimal cut off point provide a value.
  For example: what is the optimal ARBOC score is XXXX
  Please provide an AUC for delta creat alone
• Supplementary Figure 2: It would be extra nice to have a box to report the sensitivity and specificity and AUC of a creat change > 1mcmol/L/hr in isolation ......and place it above the blue bit for comparison with the full model.

• remove APACHE from model
• stages 2 and 3 aki as outcome
• RRT as outcome (likely not feasible as only 6 pts)
• gender in model
• mechanically ventilated in model (Mecvenadm)
• ABG data only vs. ABG + UEC creatinines
• redo all the pt characteristics / demographics as 3 additional pts excluded b/c they didn't have enough cr measurements (387 -> 384)

How can I access the data source? Sincerely.

• Should we add a columns with NPV, PPV, +ve LR, -ve LR sensitivity and specificity ...or at least PPV and NPV?

Need to analyse AKI stage 2 and 3 and rerun all analysis

For captions and variables within a figure, use Helvetica (or Arial) font, if possible, in upper and lower case letters.

https://journals.lww.com/ccmjournal/Documents/CCM_Info_for_Auths.pdf

Example: https://r-pkgs.org/intro.html#intro-colophon

03_screening_log

  mutate(
    Total_rows = n(),
    duplicates = Total_admissions != Total_rows,
    Event      = if_else(is.na(Pt_Study_no_olig), 0, 1),
    Event      = if_else(is.na(Pt_Study_no_crch), 0, 2) + Event
    ) %>%

• Change to Epis_cr_change
• Rerun to check
• Change to factor

Maximum of 8 total

• Figure of flow diagram showing number of pts and inclusion/exclusion
• Figure showing change in creatinine and AKI development timeframe (time vs creatinine level)
• Figure showing heatmap of creatinine changes and AKI
• Table of pts demographics (that met the inclusion criteria)
• Table of cr only as predictor
• Table of multivariable model