StatCrunch logo (home)

Report Properties
Thumbnail:

from Flickr
Created: Dec 9, 2017
Share: yes
Views: 387
Tags:
 
Results in this report
 
Data sets in this report
None
 
Need help?
To copy selected text, right click to Copy or choose the Copy option under your browser's Edit menu. Text copied in this manner can be pasted directly into most documents with formatting maintained.
To copy selected graphs, right click on the graph to Copy. When pasting into a document, make sure to paste the graph content rather than a link to the graph. For example, to paste in MS Word choose Edit > Paste Special, and select the Device Independent Bitmap option.
You can now also Mail results and reports. The email may contain a simple link to the StatCrunch site or the complete output with data and graphics attached. In addition to being a great way to deliver output to someone else, this is also a great way to save your own hard copy. To try it out, simply click on the Mail link.
Week 14- Multiple Regression Weight Recorded for GIRLS REVISION
Mail   Print   Twitter   Facebook

Introduction: Medicine dosage calculations are very important in terms of patient care and treatment. Data can be inappropriately recorded and errors can be made. This study involves a children’s hospital and their policy to institute a weight verification procedure based on the patient’s medical record. The protocol involves boundaries for weight based medical records. Any weights recorded outside these boundaries would be flagged for follow up prior to medicine distribution.

Statistical Methods: The research model involves regression models for patient weights. They will be used to develop prediction protocols for each patient. Recorded weights that fall outside of the boundaries will be flagged for follow up procedures per protocol. The following variables are identified: most recent weight (detailed in pounds), time since most recent weight (detailed in weeks), current recorded weight, and age (in years). Models will differ based on age ranges and gender; here we will look at the model for 9 to 15 year old girls.

Statistical Analysis: The first thing we must do in multiple linear regression, is to determine if there is multicollinearity. We do this by checking a pairwise correlation matrix. Since we are looking for a predictive model multicollinearity is not a big concern. 

 

Next, we need to look at our residual plots to ensure that assumptions are met. No issues with assumptions are found as evidenced by the random plot pattern.

 

Next we will look at the best model needed. We have 4 variables, so we can use a backward elimination to select the best model here.During this process we see that it removed time and pweight. 

 

Conclusion: We are left to analyze height and weight as predictors. We can see that with 95% confidence we can predict that 0.11 to 0.16 kg difference in overall weight recorded. 

Recommendations for future studies:For future studies we may have a larger sample size and longer periods of time to check on proper procedure for weight recording. We should also a change in medication errors and dosage miscalulations based on this change. 

 

 

Result 1: Correlation matrix week 14   [Info]
Correlation matrix:
AgePWeightHeightTime
PWeight0.80532188
Height0.937555710.76144183
Time0.053793257-0.121980640.061227225
Weight0.828423630.978619320.785726190.028771747

Result 2: QQ Plot week 14   [Info]
Right click to copy

Result 3: Multiple Linear Regression week 14   [Info]
Right click to copy

Result 4: Multiple Linear Regression week 14 revised   [Info]
Multiple linear regression results:
Dependent Variable: Weight
Independent Variable(s): Age, PWeight, Time, Height

Backward elimination:
P-value to leave: 0.1
StepVariableActionP-valueRMSER-squaredR-squared (adj)
1HeightRemoved0.83681.16848610.98090.9798
2AgeRemoved0.11771.18523480.980.9792

Weight = 0.018821822 + 0.98744736 PWeight + 0.091139743 Time

Parameter estimates:
ParameterEstimateStd. Err.DF95% L. Limit95% U. Limit
Intercept0.0188218220.8576415553-1.70138931.7390329
PWeight0.987447360.019397656530.948540591.0263541
Time0.0911397430.011869338530.0673328670.11494662

Analysis of variance table for multiple regression model:
SourceDFSSMSF-statP-value
Model23643.38591821.69291296.7803<0.0001
Error5374.4534161.4047814
Total553717.8393

Summary of fit:
Root MSE: 1.1852348
R-squared: 0.98
R-squared (adjusted): 0.9792

Result 5: Multiple Linear Regression week 14 revised   [Info]
Multiple linear regression results:
Dependent Variable: Weight
Independent Variable(s): Age, PWeight, Time, Height
Weight = 9.4975961 + 0.90119609 Age + 0.89927268 PWeight + 0.085124072 Time + -0.11033612 Height

Parameter estimates:
ParameterEstimateStd. Err.DF95% L. Limit95% U. Limit
Intercept9.49759615.255971353-1.044547320.039739
Age0.901196090.34411447530.210989881.5914023
PWeight0.899272680.05068856530.797604311.0009411
Time0.0851240720.018646779530.0477233710.12252477
Height-0.110336120.05735741953-0.225380530.0047082984

Analysis of variance table for multiple regression model:
SourceDFSSMSF-statP-value
Model43587.4615896.86538277.46574<0.0001
Error53171.314353.2323463
Total573758.7759

Summary of fit:
Root MSE: 1.7978727
R-squared: 0.9544
R-squared (adjusted): 0.951
95% lower limit for individual prediction stored in new column: 95% L. Limit Ind.
95% upper limit for individual prediction stored in new column: 95% U. Limit Ind.

HTML link:
<A href="https://www.statcrunch.com/5.0/viewreport.php?reportid=74360">Week 14- Multiple Regression Weight Recorded for GIRLS REVISION </A>

Comments
Want to comment? Subscribe
Already a member? Sign in.
By barrett.green
Dec 10, 2017

Candace,
Good job on your revision. I think your ideas for future studies about proper procedure for weight recording is a great idea. I can see how confounding could happen without proper procedure.
-Barrett
By nku.katie.waters
Dec 5, 2017

Hi Candace,
Good work. A few comments:
1. When interpreting the correlation matrix, we might also note the presence of multicollinearity as most of these variables have a moderate correlation to each other. Since our goal is to find a predictive model, multicollinearity is not a big concern.

2. When conducting the MLR procedure, the dependent variable is Weight and the independent variables are Age, PWeight, Time, Height. An initial MLR result gives evidence that a model containing Age, PWeight, Time and Height is useful in estimating average weight (p-value < 0.0001, F=655.5). Based on individual p-values in the parameter estimate table, PWeight and Time are identified as predictors that will be important in estimating average weight. Backwards elimination confirms that a model considering previous weight and time are the only variables that are needed in the model.

3. Once you have the "best"model, the problem asked you to use the model to make predictions. To do this in StatCrunch, you need to type the values from the table as additional rows in the dataset. Within the regression procedure, you want to select the option for "95% confidence intervals for individual prediction". The prediction intervals should appear directly in the data sheet.

Please let me know if you have any questions.
By jacobson.gina
Dec 4, 2017

Candace,
nice report. medication dosing is so important and can really effect patients. Your report helped me to further understand this very confusing concept. I think you did a good job on the statcrunch report this week.
By barrett.green
Dec 3, 2017

Candace,
Nice job on your report. I think your ideas about looking at weight change and medication errors is very important. Young women can grow quickly and this could certainly alter dosages. It's interesting, in adults in the ICU, all of our medications that are weight-based are always calculated at the patient's initial 'admission weight.' Yet, there can be instances where patients have a lot of fluid retention and can gain 20kgs. However, meds are still based on initial weight. I wonder how this affects the effectiveness of the medications when we aren't accounting for the additional weight gain or loss during a hospital admission?
Great job!
-Barrett

Always Learning