NEWS

scicalc 0.4.2

read_file_with_hash() and write_file_with_hash() no longer pass reader/writer arguments through to the hashing step. Arguments whose names collide with digest::digest() formals (e.g. skip, length, ascii, raw, seed) were being forwarded to the hash call, so something like read_file_with_hash(path, skip = 2) printed and recorded a hash of only part of the file. The file hash now depends only on the file and algo.

read_file_with_hash() now records two hashes on the object it returns: attr(df, "file_hash") is the hash of the file on disk (the one printed when reading), and attr(df, "data_hash") is a hash of the data as read in. Both use the algo you asked for.

Categorical mapping functions are now configurable through options:
- racen() recognizes six race categories (white, black, asian, american native, pacific islander, other/multiple). Set options(scicalc.racen_config = ) to a named numeric vector to override any category's code and to add novel categories (matched before the built-ins).
- agec() and bmic() accept fully custom bands via options(scicalc.agec_config = ) / options(scicalc.bmic_config = ) — a data frame with label, min (lower bound), and code columns that replaces the FDA/WHO defaults. Bands are half-open [min, next min) with an open-ended top band.
- Any active configuration is listed in the package startup message.
Added is_missing_value(), a predicate for the missing-value sentinel (getOption("scicalc.missing_value"), default -999), e.g. dplyr::filter(df, !is_missing_value(x)).
Calculation functions now handle the missing-value sentinel in their inputs: a sentinel value is treated as missing during the computation and the sentinel is preserved in the result.
read_file_with_hash() and read_hashed_file() gained a reader argument for file types they don't natively support (e.g. reader = arrow::read_feather); write_file_with_hash() gained a matching writer argument. For a known extension the custom function is ignored unless force = TRUE.

egfr(): pass method by name; positional args are sexf, raceb, age, creat, cystc, height. Passing a method string positionally now errors instead of silently defaulting.
rfc() now honors an explicitly-supplied absolute_units; the estimator's scicalc_units attribute is used only as a fallback when absolute_units is not given.
write_csv_with_hash() and write_parquet_with_hash() were filtering ... against the wrong function's arguments (read_csv()/read_parquet() instead of write_csv()/write_parquet()), silently dropping valid writer arguments like eol or compression.

The unit-tracking attribute is now scicalc_units rather than a generic units attribute. Computed quantities still carry it (egfr()/aegfr() as "mL/min/1.73m^2" / "mL/min", bmi() "kg/m^2", bsa() "m^2", ibw()/aibw() "kg", crcl() "mL/min", convert_alb/bili/creat() "g/dL"/"mg/dL"), and rfc() reads it to infer relative-vs-absolute when absolute_units is not supplied.

The individual eGFR equation functions (ckdepi_2009_egfr(), ckdepi_2021_egfr(), ckdepi_2021_egfr_cystatin(), mdrd_egfr(), schwartz_egfr()) and BSA equation functions (dubois_bsa(), mosteller_bsa()) are deprecated. Use egfr(method = ...) and bsa(method = ...) instead. They will become internal in 0.6.0.
read_csv_with_hash(), read_parquet_with_hash(), read_sas_with_hash(), read_xpt_with_hash(), read_excel_with_hash(), and read_pzfx_with_hash() now warn on every call (previously a soft deprecation). Use read_file_with_hash().
write_csv_with_hash() and write_parquet_with_hash() are deprecated. Use write_file_with_hash() instead. They will become internal in 0.6.0.

Added aegfr() function to convert relative eGFR (mL/min/1.73m^2) to absolute eGFR (mL/min).
Added ibw() (ideal body weight) and aibw() (adjusted ideal body weight) functions.
Added bmic() (BMI category) function.
Added agec() (age category) function.
Added cor_df() function that takes a dataframe and a vector of column names to compute all pairwise correlations.
Added is_female() and is_black() helper functions for converting character sex/race to logical values.
Added unit conversion functions: convert_creat(), convert_bili(), convert_alb().
Added round_like() function to round numeric values based on explicit or inferred significant digits from a reference value. (Experimental, API may change.)

Computed quantities now include a units attribute for unit tracking:
- eGFR functions: "mL/min/1.73m^2" (relative) or "mL/min" (absolute)
- crcl(): "mL/min"
- bsa(), dubois_bsa(), mosteller_bsa(): "m^2"
- bmi(): "kg/m^2"
- ibw(), aibw(): "kg"
- convert_creat(), convert_bili(): "mg/dL"
- convert_alb(): "g/dL"
Category functions now include a category_standard attribute:
- rfc(): "FDA" or "KDIGO"
- hfc(): "NCI-ODWG"
- bmic(): "WHO"
- agec(): "FDA"
rfc() can now infer absolute_units from the input's units attribute, allowing rfc(AEGFR) to work directly without specifying units.
Updated egfr() equation default to use CKDEPI 2021 equation.
Default hash algorithm changed from md5 to blake3 for read_file_with_hash() and related functions.
Added recycling warning to all functions when inputs have different lengths.

read_csv_with_hash(), read_parquet_with_hash(), read_sas_with_hash(), read_xpt_with_hash(), read_excel_with_hash(), and read_pzfx_with_hash() are soft-deprecated in favor of read_file_with_hash() which auto-detects file type.

Fixed edge case handling for hfc when bilirubin is exactly equal to the cut-off values of 1.5 or 3 times ULN