"Association" between attributes; odds ratio; confidence interval

DocMusherDocMusher Member Posts: 236   Unicorn
Hi,

This paper deals with principally association of 2 attributes where the label is known for all examples. The data is recalculated with filters which are based on values currently considered as standard in medicine.

I do not get how this analysis could be repeated using RM. I know there is some discussion on the use of odds ratio etc.
The reason for my question is the fact I would like to analyse other "associations" of other attributes.

What is the correct terminology for the process used in this article?

Thanks
Sven


https://www.dropbox.com/s/pfh7tm1j2cp1dxd/J%20Intensive%20Care%20Med-2014-Moskowitz-0885066614530659.pdf?dl=0

Answers

  • DocMusherDocMusher Member Posts: 236   Unicorn
    Hi,
    Anyone a python script available to calculate the odds ratio and confidence interval?
    Thanks
    Sven
  • mschmitzmschmitz Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 1,877  RM Data Scientist
    Hi Sven,

    could you be a bit more specific on what "filter" you want to use? It is kind of hard for me to read those papers, because the language is just different from the papers i am used to read.

    Cheers,
    Martin
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • DocMusherDocMusher Member Posts: 236   Unicorn
    Dear Martin,
    With filters I meant data ranges used in medicine, most frequently empirically defined. For instance:  serum lactate concentration >2 mmol/L is considered lactate acidosis,  severe lactate acidosis is lactate concentration >4mmol/L, normal lactate concentrations <= 2mmol/L
    Cheers
    Sven
  • mschmitzmschmitz Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 1,877  RM Data Scientist
    So what is the difference to a usual filter examples?
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • DocMusherDocMusher Member Posts: 236   Unicorn
    Examples are filtered indeed, but based on "experience", "empirical" knowledge, it is not the process that finds the threshold in this context.


    Cheers
    Sven
  • mschmitzmschmitz Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 1,877  RM Data Scientist
    Sorry Sven i do not get what you want to do more than filter examples. I most likely get something wrong...

    Cheers,
    Martin
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • DocMusherDocMusher Member Posts: 236   Unicorn
    Hi,
    The authors of the paper report:  We report baseline patient characteristics and laboratory data by quartiles. Sequential, multivariate models were created to define the association between magnesium concentrations, assessed as quartiles  and  as  a  continuous  variable....For  all  analyses  of cohort quartiles, the third magnesium quartile was used as reference.  This  reference  value  was  chosen  as  it  most  closely reflects  normal  magnesium  concentrations.... Separate  regressions  for  lactic  acidosis  (>2  mmol/L)  and severe lactic acidosis (>4 mmol/L) were performed. In unadjusted analysis (Table 2),the lowest magnesium quartile had an 85% increased odds of mild  lactic  acidosis  (odds  ratio  [OR]  1.85,  95% confidence interval [95%CI] 1.64-2.08,P< .001) compared to patients in the reference quartile. This relationship remained highly significant after adjustment (OR 1.71, 95%CI 1.51-1.94, Pvalue<.001).

    My question is short: with the exampleset available, how could I calculate the OR and confidence interval?
    Thanks
Sign In or Register to comment.