Figure Stack

Latest Figures

New To Do

1.Distribution of GD diagnosis time

GD diagnosis time vs birth time

GD diagnosis with other pregnancy complications

You can probably produce a single dataframe to have all this info. Having a bunch of patients with diagnosis time, birth time, and other complications (maybe flags for certain complications) would suffice

2. Indicative Diseases With obvious codes

Need to put this as supplementary. So dont rewrite what there is now

3. Perturbation Analysis

Increase of raw risk with one extra endocrine code, and one extra infection, asa function of time (in the training period)

To Do

Update results with BEST Auc

Please stop hobbling the results. We need to put our best (but correct) foot forward

Generate Precision-Recal Curves

The following snippets are helpful.

Please generate PRC for each time points that you ahev used in the rest of the analysis

def selROC(df,fpr=.1,tpr=None,th=None,eps=0.01,p=0.017,ppv=False):
    row=df[df.fpr.between(fpr-eps,fpr+eps)].max()
    if ppv:
        ppv=row.tpr/(row.tpr+row.fpr*((1./p - 1)))
        return row,ppv
    else:
        return row.tpr
def getPPV(row,RHO):
    return row.tpr/(row.tpr + row.fpr*(1./RHO -1))

Short labels for feature names

Please write short, undertstandable labels for fetaure names, that we can use in the ylabels of Fig. 1C. You can also have slightly longer description of what things like mean position etc. means that we can add to teh figure caption, and which makes sense when someone reads the caption only.

Confidence Bounds for AUC

Please draw samples from the population, compute AUCs for the different time points, and compute upper and lower confidence bounds using the formula: $$\hat{y} = y \pm \frac{2.63\sigma}{\sqrt{n}} $$

The header of the resultant csv file is the following:

lb,v,ub,time,gender,dataset

where dataset currently is just TRUVEN, and time here is one of 0,-1,-2,-4

Generate Publisher Files for Postpartem