1.

Clinical records (e.g. SDC electronic patient records)

- Complete history of patients:
  . HbA1c
  . lipids
  . blood pressure
  ...

- Information on:
  *dates* of measurement
  *date* of diagnosis
  *date* of birth

- Intervals between visits depend on patients' status


2.

Clinical registers (e.g. Danish Adult Diabetes database)

- Data collection (recording) at fixed intervals
  (once a year, e.g.)

- Clinical data on individuals

- Data collection independent of patients' clinical status w.r.t.
  . HbA1c
  . lipids

- Missing data:
  . a patient was not seen for an entire year
  . a patient has moved
  . a patient died (but was not recorded as such)


3.

Population level registers (e.g. National Diabetes Register)

- (cl)Aims to cover the entire population

- Limited information on each patient:
  . date of birth
  . date of diagnosis
  . date of death
  . sex
  ...

- Monitoring of:
  . DM occurrence (incidence rates)
  . prevalence of DM
  . mortality of DM patients

- Important for:
  . long term follow-up 
  . no patient drop-out



[Jeg lægger tegninger til de følgende i http://BendixCarstensen.com/SDC/STAR]

4. Prevalence of DM in Denmark 1995--2012 by sex
  (tegning)

5. SMR 
   - the ratio of mortality in DM patients to the mortality in the
   non-diabetic part of the population.
   (tegning)


6. Use of clinical registers

   - Recall: Registers that collect clinical information on patients at
     regular intervals

   - How many % attain a HbA1c < 7% (xx mmol/mol) ?

   - How many % attended eye screening in the last year ?

   - How frequent are complications in different ethnicities?


7. Complication frequency in Danish DM patients by ethnicity:
 
   (JkOy's tegninger)


8. Renal complications and CVD in SDC T1 patients

   - patients with DN (diabetic nephropathy)

   - occurrence of
     . ESRD (end stage renal disease: dialysis or transplant)
     . death

   - how do rates depend on clinical parameters?

   - how is long-term outcome dependent on clinical status?


 9. Renal complications (2): Tegning af kasserne (DN1-boxes-tp1.pdf)
10. Renal complications (3): Forest plot for kliniske variable (DN1-xforestcol1.pdf)
11. Renal complications (4): Probability plot (DN1-pr55.pdf, DN1-pr45.pdf)


12. Requirement for analysis of clinical records
    
    - Well defined patient population (what is DN, CVD, ESRD)

    - Well defined research question:
      . effect of clinical variables
      . on rates
      . on long-term outcome

   - Only possible through close collaboration between 
     . clinical researchers - what is relevant, what is available, what
       is reliable
     . statistician - what is possible, what is relevant, what data is
       needed 

   - The project took many hours of joint discussion to get the boxes
     right, and the hypotheses properly hammered out.
  
    
13-20 eksisterende slides:

Majkens projekt (2 slides)
T1 mortalitet (3 slides)
Foot ulcers (3 slides)


14. Use of clinical records: requirements, *DATA*
    
   - Well defined patient population:
     . Start of attendance
     . End of attendance - who is no longer affiliated with the
       clinic - otherwise we run the risk of counting persons who dies
       without our knowledge
   - Well defined (time-consistent) variable definitions
     . Measurement methods are the same over time?
     . Is the indication for measurement the same over time; this
       influences the actually obtained measurement values.


14. Use of clinical records: requirements, *ANALYSIS*

   - Outcome definition (response, dependent variable):
     . Death
     . HbA1c
     . Healing of foot ulcer

   - Explanatory variables (predictors, independent variables) 
     . sex, age
     . calendar time
     . clinical measurements
     . treatment

   - Note: Using treatment as explanatory variable induces (almost
     invariably) *confounding by indication*:

     Patients are treated for a reason: 
     - the more treatment the worse the outcome, because
     - treatment is a proxy for clinical status (beyond measurable variables)


15. Use of clinical records: requirements, *STATISTICS*

    - Continuous outcomes:
      . HbA1C
      . lipids
      . GFR
      ...
      - require repeated measures models
        (aka. mixed models, random effects models)

    - Event type outcome:
      . death
      . ESRD
      . retinopathy
      - require survival-type analysis:
        . death - survival analysis
        . all other: competing risks or multistate models


16. Clinical records, use of databases:

    - Describe data:
      . WHO
      . WHAT
      . WHEN
      . (WHY)

    - Describe hypothesis or research question
      . WHAT
      . depend on WHAT
      . and in particular HOW MUCH
      
    - Always specify research question in QUANTITATIVE terms, never
      "is there an effect of...".

      There is one, but maybe so small that we do not bother.