feat: Compute d2:zScore functions with WHO LMS method [DHIS2-22201]

[enricocolasante]

Compute d2:zScore* functions with the WHO LMS method DHIS2-22201

Problem

The z-score functions looked up 7 SD values per row (−3 SD … +3 SD, rounded to 1 decimal) and interpolated linearly between them. This had several issues:

• Inaccurate results. Results differed from WHO Anthro by up to ~0.1, because the SD values are rounded and WHO's curve between SD points isn't a straight line.
• Values beyond ±3 SD were always ±3.5. Anthro keeps counting past ±3.
• d2:zScoreWFH used the wrong reference between 87 and 110 cm. It used weight-for-length values (meant for children measured lying down) up to 110 cm. When age is unknown, the usual convention is to switch to weight-for-height at 87 cm. Children 87–110 cm tall were compared against a reference 0.1–0.4 kg too light, so their z-scores came out too high.
• Data errors:
  • the female WFH row at 112 cm was corrupted (an LMS parameter in place of −3 SD, with the other columns shifted)
  • boys' weight-for-age had no month 28, so d2:zScoreWFA(28, …, "m") threw an error
  • height-for-age at 24 months used the length table for girls but the height table for boys

Changes

Calculation (ZScore)

• The formula: z-scores are computed from WHO's L, M and S parameters, the same way WHO Anthro does it: z = ((x / M)^L − 1) / (L · S).
• Beyond ±3 SD: WHO's rule applies, so the z-score keeps counting in steps of the distance between the 2nd and 3rd SD instead of being capped at ±3.5.
• Precision: calculations use Double (no float noise), results are rounded to 2 decimals, and 0.0 is returned instead of -0.0.

Tables (ZScoreTable): each row holds the WHO L, M, S parameters instead of 7 rounded SD values.

• WFH: every 0.1 cm from 45.0 to 120.0 cm, with weight-for-length below 87 cm and weight-for-height from 87 cm.
  • Heights between two rows (e.g. 112.37 cm) blend L, M and S of the neighbouring rows, like Anthro.
  • Between 86.9 and 87.0 cm the length curve is extended from its last two rows, so the blend never mixes the two tables.
• WFA: WHO monthly parameters for 0–60 months.
• HFA: WHO monthly parameters for 0–60 months, with length-for-age below 24 months and height-for-age from 24 months for both sexes.
• Cleanup: the old SD-table code is removed, including the public ZScoreTable.Entry (see the API dump).

Behaviour changes

• Most results shift slightly, by up to ~0.1, and the corrected 87–110 cm WFH range shifts by up to ~0.4. Former whole numbers now usually have decimals (e.g. 0.02 instead of 0.0).
• Values beyond ±3 SD are now real numbers (e.g. −4.71) instead of ±3.5. Program rules that check for exactly ±3.5 need updating.
• WFH now accepts any height between 45 and 120 cm, not only values on a 0.1 cm step.
• WFA/HFA still take whole months: fractional ages throw, as before.
• API: ZScoreTable.Entry is removed and ZScoreTable.Lms is added.

Validation

• Matches WHO Anthro to the hundredth for the reference cases. These include children over 2 measured standing, and infants (for the infants, Anthro was set to "measured standing", so the tests add its 0.7 cm adjustment).
• Data sources: WHO growth standards LMS files (wflanthro, wfhanthro), as packaged with the WHO anthro R package, and WHO's monthly weight-for-age and length/height-for-age tables.
• Tests cover:
  • the 87 cm and 24-month switches
  • heights between 0.1 cm rows
  • values beyond ±3 SD
  • the table limits
  • unknown parameters
  • the Anthro reference cases

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
99.9% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud