topic Re: overlapping variables in a composite index in Discussions

overlapping variables in a composite index

Sandeep123 — Sun, 11 Jun 2023 11:12:28 GMT

I have a data set of patients of which I have created a composite index of critical illness. This composite includes patients who died, who had ventilator requirement, who had septic shock , required dialysis et cetera (about total 16 categories all nominal yes/no).

Reviewer want to know how much of these categories contributed to the composite. There are a total 600 patients who met the criteria of critical illness based on my definition.

I can provide information like the number of patients who died (n=xx), ventilator (n=yy).. All of them would be in the composite.

The sum of all these patients would be more than 600 because all of them are overlapping with the other.

Is there a way to solve this and maybe create a graph or diagram showing the overlap . I'm thinking of the bubble plot but there might be another way of doing this.

Any help much appreciated

Re: overlapping variables in a composite index

ih — Mon, 29 Mar 2021 12:00:31 GMT

I'm not sure I fully understand your question so I am starting really simple, could you show number of unique patients with each condition? In this example every patient was tested for every condition so the total is the same, thus I would remove the 'does not apply' bars.

This script will reproduce the graph above:

Names default to here(1);

dt = Open("$Sample_data/Nicardipine.jmp");

dtTab = (Data Table( "Nicardipine" ) << Tabulate(
	Add Table(
		Column Table( Category Table, Statistics( Name( "% of Total" ), N ) ),
		Row Table(
			Columns by Categories(
				:Patient Died Flag,
				:Anticonvulsants Flag,
				:Antiemetics or Phenothiazines Flag,
				:Antifibrinolytics Flag,
				:Antihypertensives Flag,
				:Blood Transfusion Flag,
				:Central Venous Pressure Monitoring Flag,
				:Induced Hypertension Flag,
				:Intentional Hypervolemia Flag,
				:Intentional Hemodilution Flag,
				:Low Molecular Weight Dextran Flag,
				:Mannitol Flag,
				:Steroids Flag,
				:Swan Ganz Monitoring Flag,
				:Vasopressors Flag
			)
		)
	)
)) << Make Into Data Table;
dtTab:"N(N)"n << Set Name("Applies");
dtTab:"N(Y)"n << Set Name("Does Not Apply");

dtTab << Graph Builder(
	Size( 534, 464 ),
	Show Control Panel( 0 ),
	Variables(
		X(
			:Columns by Categories,
			Order By( :Applies, Descending, Order Statistic( "Mean" ) )
		),
		Y( :Applies ),
		Y( :Does Not Apply, Position( 1 ) )
	),
	Elements( Bar( X, Y( 1 ), Y( 2 ), Legend( 5 ), Bar Style( "Stacked" ) ) ),
	SendToReport(
		Dispatch(
			{},
			"400",
			ScaleBox,
			{Legend Model(
				5,
				Properties( 0, {Fill Color( 27 )}, Item ID( "Applies", 1 ) ),
				Properties( 1, {Fill Color( 32 )}, Item ID( "Does Not Apply", 1 ) )
			)}
		),
		Dispatch(
			{},
			"graph title",
			TextEditBox,
			{Set Text( "Patients with Conditions" )}
		),
		Dispatch( {}, "X title", TextEditBox, {Set Text( "Conditions" )} ),
		Dispatch( {}, "Y title", TextEditBox, {Set Text( "Count" )} )
	)
);

Re: overlapping variables in a composite index

dale_lehman — Mon, 29 Mar 2021 12:17:10 GMT

I would think the most immediate measure would be the total number of the 16 indicators that each patient had. To investigate the contribution of each indicator to the composite, Principal Component Analysis comes to mind (although there may be a better method with nominal variables, such as Multiple Correspondence Analysis).

Re: overlapping variables in a composite index

Sandeep123 — Mon, 29 Mar 2021 14:55:31 GMT

I used predictor screening platform in JMP. With Composite index yes/no as Y variable and added all the variables as X.

It gave me ranking of all the variables as they contributed to the index along with their percentage contribution. This will work for my purpose, not 100% sure it is the right way though as the variables are components of the composite