Regsitration

[marrpesce]

Hi Andrea,
I introduced changes to improve coherence and readability across the analysis.

0_utility_functions.R
Defines reusable helper functions used across the analysis scripts.

1_derive_key_variables.R
Builds the core processed dataset used in all downstream analyses. It derives key variables including death source (ONS/TPP/Both), inclusion flags, reference death date, registration status and timing, TPP-coded death indicators, date differences, and demographic variables.
Outputs: death_registration_processed.csv.gz (main dataset), death_registration_processed_skim.txt, table_source_raw.csv

2_implausible_death_dates.R
Assesses implausible death dates separately for ONS and TPP (e.g. before birth, outside study period). This acts as a data quality check.
Output: death_implausible_source.csv

3_registration_at_death.R
Describes registration status at the time of death and timing of registration relative to death. Restricted to valid death dates.
Outputs: registration_status_source.csv, reg_start_timing_source.csv, reg_end_timing_source.csv

4_death_source_comparison.R
Compares where deaths are recorded (ONS only, TPP only, both), overall and by subgroup. Includes a sensitivity analysis incorporating TPP-coded deaths.
Outputs: table_death_source.csv , table_death_source_25_26.csv , tpp_death_code_or_date.csv , table_death_source_overall_any_tpp.csv

5_sources_date_agreement.R
Assesses agreement between ONS and TPP death dates among individuals with records in both sources, grouping differences into meaningful categories.
Output: table_ons_tpp_dates_diff.csv

6_variation_source_by_practice.R
Quantifies variation across GP practices in the proportion of deaths recorded only in ONS, summarised using practice-level percentiles by year.
Output: table_practice_percentiles.csv

Post-release script
Scripts in the /post_release folder generate additional outputs derived from the main analysis tables after disclosure control, for tables / visualisation.

[alschaffer]

Technically this should be days <= -31 ~ "<=-31". Just a labelling thing so not a big deal as long as you interpret it correctly.

[marrpesce]

Thank you, I changed it to -31+

[alschaffer]

I am curious about the decision to extract last recorded event, as opposed to first. I would have assumed the first coded event was most relevant

[marrpesce]

Thank you, changed to .first_for_patient()

[alschaffer]

Would this work:

has_possible_age = (
((age_at_death >= 0) & (age_at_death < 110))
& (date_of_birth >= ref_death_date)
)

?

B/c age is calculated in whole years - if we include babies born in the year of their death (like you have) that will include people whose date of death is before their date of birth, and is a likely error.

[marrpesce]

Thank you, changed to >= 0

[alschaffer]

I think you mean has_non_disclosive_sex

[marrpesce]

Thank you! I fixed this

[alschaffer]

it would be better to sum first, then apply rounding - if you sum rounded values, that will introduce additional uncertainty - probably not a big deal in the grand scheme of things but it is better practice.

[marrpesce]

Modified to compute yearly totals first and apply rounding after summing, while keeping subgroup counts rounded before calculating percentages

     group_by(death_date_ref_year, death_source) |>
  summarise(
    total = n(), 
    .groups = "drop"
  )|>
  group_by(death_date_ref_year) |>
  mutate(
    total_year = rounding(sum(total, na.rm = TRUE)),    
    total = rounding(total),
        perc = total / total_year * 100          
  )