CancerRegistryOfNorway / nordcansurvival

Other
0 stars 0 forks source link

reporting CI for RS > 100% #39

Closed BjarteAAGNES closed 2 years ago

BjarteAAGNES commented 2 years ago

ref emails :

nonsense CI for RS > 100%: 101.0 [101.0-101.0] (example FI lip) "useless" estimate 100.0 [0-100.0]

  1. Propose fix for current display of results
  2. Propose change to next production

FOLLOW-UP WEDNESDAY 25 AUGUST

BjarteAAGNES commented 2 years ago

The CI is the std CI of stnet (compute ci after a log-log transformation.), but should be replaced with (what probably should be default) by using the option cilog which compute CI on log scale.

The survival_statistics.ado will be updated adding the option option cilog to the current:

ncstnet , /// lifetable("lifetable'") /// outfile("tmp'/entityentity'") /// by(by') ///
breaks(breaks') /// iweight(iweight') /// standstrata(`standstrata')

BjarteAAGNES commented 2 years ago

COMMENT: The reason for this error is probably that the code is based on Cancer in Norway, and at some point we made a special version of stnet implementing the log scale CI before the official stnet was updated with this possibility, but then as a option. A note is given to the CIN team to review this and possibly change to the official stnet using the option cilog.

BjarteAAGNES commented 2 years ago

Mulig denne har blitt liggende eller er besvart via epost:

101.0 [101.0-101.0]

bør endres til

101.0 [NA-NA]

CotterpinDoozer commented 2 years ago

Vi kommer til å fikse dette i selve koden, ikke sant? Slik at det ikke skjer i fremtiden når man bruker ny versjon av nordcan.R?

Men MÅ vi fikse det i visningen i NORDCAN nå? Hvordan skal IARC evt. finne de som skal få NA i stedet for den verdien som står der nå?

BjarteAAGNES commented 2 years ago
  1. Dette vil bli rettet. Ref forklaring i kommentarer over.
  2. Hvis det ikke er hensiktsmessig å rette kan en kommentar legges inn av meg i dokumentasjonen på github som gir en forklaring. Konfidensintervall hvor nedre og øvre grense er identiske åpenbart nonsens. En eventuell korrigering fra IARC er en søk erstatt, hvor man bruker regulære uttrykk, og et Stata eksempel kan være :
    
    clear
    input str25 eci 
    "101.0 [101.0-101.0]"
    "101.1 [101.1-101.1]"
    "101.2 [101.2-101.2]"
    end

replace eci = ustrregexrf(eci, "\s[1\d\d.\d-1\d\d.\d]", " [NA-NA]")

. list , clean

             eci  
  1. 101.0 [NA-NA]
  2. 101.1 [NA-NA]
  3. 101.2 [NA-NA]
CotterpinDoozer commented 2 years ago

I will check with IARC if they can fix this by search and replace now, or if we need to just document in GitHub until this is fixed in the norcan.R.

BjarteAAGNES commented 2 years ago

https://github.com/CancerRegistryOfNorway/nordcansurvival/commit/50bad377c5d03ba76e306aeb588d799f3732d7f0