Validation of registers Jonas F Ludvigsson Gothenburg March 3, 2015 Do we measure what we want to measure?
Short about myself Jonas (F) Ludvigsson Sweden s prime castle? Pediatrician/professor 50/50 Football, guidebooks, Tintin Three boys Hannes, Elias and Josef ( HEJ )
Acknowledgement Thanks to Henrik T Sørensen Int J Epi, 1996. A framework for evaluation
Is this validation?
Key elements of medical databases The population Medical events, diseases and other data Organization of data collection and management
Limitations (national registers) Few lifestyle factors If lifestyle factors: how recorded, when? Lack of data on confounders Lack of data on disease severity, location, treatment
Our aim 800 600 Reduce systematic and random errors 400 200 0 Nov Dec Jan Feb Occurrence over time: GE in Örebro
Our aim (2) Reduce systematic and random errors Major epidemics Not congenital malformations
Our aim (3) Relative risk Elfström, JNCI, 2011
Validation > lower power ECG: ejection fraction <40% n=199 >166 ECG: left ventric. diastolic diameter<32 mm/m2 Angiography without >50% stenosis No ischemic heart disease No valvular failure No severe hypertension n=69 p=0.052
Estimate disease frequency Validate only the cases > lower prevalence n=40 -> n=20 Prognosis If severe cases fulfil more criteria [or available in Patient Register] Poorer prognosis
Completeness (1) Merge several datasources Completeness: numbers in data 1/numbers in all sources A diagnosis failing to account for all cases Anaphylaxia -> Bee sting
Completeness (2) No Merge Compare aggregated data his prevalence is similar to that in a recent Finnish study This prevalence is consistent with national data
Completeness (3) Not always a problem Case Control exposed unexposed (as long as case identification is unrelated to exposure)
Patient discharge data Variation in coding Errors in coding incompleteness in coding Poor specificity Errors and variation in clinical diagnosis
Size of database False-positive findings? Clinical significance? Approach: Not only the mothers but also the fathers
Registration period (1) 1964 1990 Ludvigsson et al. BMC Public Health 2011, 11:450 1964 Psychiatry 2006
Registration period (2) Short follow-up: not a good idea for cancer research ERC application - I failed. Examining the influence of HP eradication on cancer, brilliant idea but not a perfect database
Registration period (3) Changes in ICD, changes in content
Traditional patient chart validation (1) Think hard - has anyone done the job for you?
Traditional patient chart validation (2) Reduce systematic and random errors Ludvigsson et al. BMC Public Health 2011, 11:450: The Patient Registry
Traditional patient chart validation (3) Ethics approval The patient charts Socialstyrelsen (they may say no ) (currently - psychiatric diagnoses)
The right numbers 1. 555+K50 Patientregistret Sluten n=33574 2. 556+K51 Patientregistret Sluten n=42530 3. 555+K50 Patientregistret Öppen n=38502 Minst två diagnoser av ovanstående koder 4. 556+K51 Patientregistret Öppen n=59684 5. 555+K50 Patientregistret Sluten-Åtgärder n=159154 6. 556+K51 Patientregistret Öppen-Åtgärder n=12573 Kör dessa patienter mot SWIBREG Välj 200 patienter som också finns i SWIBREG Patienter som finns i SWIBREG men INTE finns i Patientregistret:! välj 263 patienter ur SWIBREG som inte har någon kod för 555/556/K50/K51 välj 180 patienter ur SWIBREG som opererats men saknar relevant åtgärdskod i Patientreg. Välj 50 pat Välj 50 pat Välj 50 pat Kör dessa patienter mot SWIBREG Välj 50 pat 100 pat Spara som K50slutenatgard 100 pat Spara som K51slutenatgard n=25 ej i Swibreg sparas: K50sluten1 25 återfinns i SWIBREG Spara som K50sluten2 Lägg till 30 patienter som finns i både SWIBREG och Patientregistret Spara som K50sluten3 n=28 sparas: K51sluten1 22 återfinns i SWIBREG Spara som K51sluten2 Vi har nu 25 patienter som finns i både SWIBREG och Patientregistret. Jag vill komplettera upp så att jag får 55 patienter som både funnits i SWIBREG och Patientregistret n=17 sparas: K50öppen1 33 återfinns i SWIBREG Spara som K50oppen2 Lägg till 33 patienter som finns i både SWIBREG och Patientregistret Spara som K50sluten3 Lägg till 22 patienter som finns i både SWIBREG och Patientregistret Spara som K50sluten3 n=19 sparas: K51öppen1 31 återfinns i SWIBREG Spara som K51oppen2 Turkosa rutor Gröna rutor Lila rutor! alla antal är påhittade Lägg till 24 patienter som finns i både SWIBREG och Patientregistret Spara som K50sluten3 GRÖNT + LILA! Den här kompletteringen innebär att jag får 60 patienter som varit i både SWIBREG och Patientregistret för varje grupp.! Det motsvarar 220 patienter - alltså det jag bett om tillstånd för
Getting the patient charts Team up with a clinician Make your case Make it easy Time periods Data you want? (give an example) Never give up Send out reminders Add phone number Admission Lab data Daganteckning Surgery Discharge Histopathology Radiology kl 09.45: Jag ringer från Regionarkivet i Gbg
Sometimes you have to compromise
Or perhaps you do not want the patient charts? Request very little data Sensitive data (psychiatry, gynaecology etc) Head of department objects Q1: yes/no Q2: yes/no Q3: yes/no
Keep track of the charts Organized according to personal identity number
Data extraction - software Reduce systematic and random errors
Data extraction - software Reduce systematic and random errors
Data extraction - software
Data extraction - software
Is missing negative Missing data=negative?
What do the journals want? Clinical Epidemiology publicerer rigtig gerne valideringsartikler. Vi lægger bl.a. vægt på, at der kommer kvantitative estimater for mindst en prædiktiv positiv værdi og helst også for completeness. Henrik T Sørensen, Editor
The future Reduce systematic and random errors http://record-statement.org
The End jonasludvigsson@yahoo.com Look, no dot
Linking data to find cases