What to Include in the Outcome Metrics for Depression Treatment?--Taking a Look at the 50% Improvements in Different Aspects of Depression

Sin-Ying (Alina) Lin · August 29, 2022

Descriptive OutcomeMetrics

When discussing treatment outcomes, we tend to focus solely on symptom reduction. We typically assume that other aspects of improvements always follow core symptom reduction. However, researchers rarely examine the overlaps between core symptom reduction and different types of outcome improvements.

Here, we examined whether other types of outcome improvements follow core symptom reduction (here, depressive symptoms) in patients with depression after completing a partial hospital program.

We define clinically significant improvement as 50% of improvement in one measured domain.

We examined the overlaps of individuals with 50% improvement in reduction of depressive symptoms with 50% of improvements in:

Reduction of non-depressive symptoms
Positive mental health
Functioning
Well-being

We also compared the global patient rating of overall improvement in each response group:

Responders to both domains (i.e., reduction of depressive symptoms and another measured domain)
Responders to depressive symptom reduction only
Responders to the other domain only
Non-responders to both domains

Load Packages

import pandas as pd
from dfply import *
import numpy as np
from plotnine import *
import statsmodels.api as sm
from statsmodels.formula.api import ols
from scipy.stats import ttest_ind

Customize Notebook Displays

pd.options.display.max_seq_items=300
pd.options.mode.chained_assignment = None  # default='warn'
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.simplefilter(action="ignore", category=SettingWithCopyWarning)

Load Datasets

#RDQ: Different Types of Treatment Outcomes
rdqPre = pd.read_spss('../Data/PrePostMeasures/RDQPre_1.sav') #Pre-treatment
rdqPost = pd.read_spss('../Data/PrePostMeasures/RDQPost_1.sav') #Post-treatment

#Discharge Package: Patient perceived satisfaction & Improvement 
satOld = pd.read_spss('../Data/PHPSatisfaction/EndofTreatmentSurveyData/EOT Verfication_1.sav')#Old data before 2020
satNew = pd.read_spss('../Data/DischargePacket/2022-06-06 dc.sav') #New data after 2020
#We swtiched the platform for data collection in 2020 so we have two separate datasets

#Diagnosis
dx = pd.read_spss('../Data/DemosDx/Diagnosis_1.sav') 

#Demographics
demo = pd.read_spss('../Data/DemosDx/Demographics Form_1.sav')

Preprocess Data

#Change ID type from float to object
rdqPre.ID1 = rdqPre.ID1.astype('object') 
rdqPost.ID1 = rdqPost.ID1.astype('object') 

#Replace text with level orders 
rdqPre = rdqPre.replace(
    {'not at all or rarely true': 0, 'sometimes true': 1, 'often or almost always true': 2})
rdqPost = rdqPost.replace(
    {'not at all or rarely true': 0, 'sometimes true': 1, 'often or almost always true': 2})

#Convert reverse items in RDQ
for i in [29,30,48,49,50,51,52] :
    colname1 = 'rdqpre_' + str(i) + '_1'
    colname2 = 'rdqpost_' + str(i) + '_1'
    rdqPre[colname1] = abs(rdqPre[colname1].astype('float') - 2)
    rdqPost[colname2] = abs(rdqPost[colname2].astype('float') - 2)
    
#Rename variables to ensure consistency between the old and new datasets of the global rating of improvement
satGlobalOld = satOld.loc[:,['PHP_ID_1', 'IMPRV_1']].rename(columns = {'PHP_ID_1': 'ID1','IMPRV_1': 'imprv1'})
satGlobalNew = satNew.loc[:,['id1', 'imprv_1']].rename(columns = {'id1': 'ID1','imprv_1': 'imprv1'})

#Merge old and new discharge package (i.e., overall satisfaction/improvement) datasets
imprv = pd.concat([satGlobalOld, satGlobalNew],axis = 0) #Use concact to merge data by columns
imprv = satGlobal[satGlobal.ID1 != 0] #Remove invalid IDs
imprv.ID1 = imprv.ID1.astype('object') #Change ID type from float to object

Create the five domains of improvements:

Reduction in depressive symptoms
Reduction in non-depressive symptoms
Positive mental health
Functioning
Well-being

rdqPre['pre_dsym'] = rdqPre.iloc[:,1:15].mean(axis = 1) #Pre-treatment Symptom Reduction in Depressive Symptoms
rdqPre['pre_ndsym'] = rdqPre.iloc[:,15:26].mean(axis = 1) #Pre-treatment Symptom Reduction in Non-Depressive Symptoms
rdqPre['pre_cope'] = rdqPre.iloc[:,26:31].mean(axis = 1) #Pre-treatment Coping
rdqPre['pre_pmh'] = rdqPre.iloc[:,31:43].mean(axis = 1) #Pre-treatment Positive Mental Health
rdqPre['pre_fun'] = rdqPre.iloc[:,43:53].mean(axis = 1) #Pre-treatment Functioning
rdqPre['pre_well'] = rdqPre.iloc[:,53:61].mean(axis = 1) #Pre-treatment Well-being

rdqPost['post_dsym'] = rdqPost.iloc[:,1:15].mean(axis = 1) #Post-treatment Symptom Reduction in Depressive Symptoms
rdqPost['post_ndsym'] = rdqPost.iloc[:,15:26].mean(axis = 1) #Post-treatment Symptom Reduction in Non-Depressive Symptoms
rdqPost['post_cope'] = rdqPost.iloc[:,26:31].mean(axis = 1)#Post-treatment Coping
rdqPost['post_pmh'] = rdqPost.iloc[:,31:43].mean(axis = 1) #Post-treatment Positive Mental Health
rdqPost['post_fun'] = rdqPost.iloc[:,43:53].mean(axis = 1) #Post-treatment Functioning
rdqPost['post_well'] = rdqPost.iloc[:,53:61].mean(axis = 1)#Post-treatment Well-being

Here we selected two groups of samples.

One group contains all patients with a diagnosis of major depressive disorder (MDD). In contrast, the other group comprises only patients with MDD as their primary diagnosis (i.e., they primarily came here for depression).

#Find patients with MDD
def findMDDAll(df: pd.DataFrame) -> pd.DataFrame:
    return df[df.mddsnp_1.str.contains('Curr|Prin') | df.mddrnp_1.str.contains('Curr|Prin')|
        df.mddsp_1.str.contains('Curr|Prin')| df.mddrp_1.str.contains('Curr|Prin')]

#Find patients with MDD as their primary diagnosis
def findMDDCurrPrin(df: pd.DataFrame) -> pd.DataFrame:
    return df[df.mddsnp_1.str.contains('Curr, Prin') | df.mddrnp_1.str.contains('Curr, Prin')|
        df.mddsp_1.str.contains('Curr, Prin')| df.mddrp_1.str.contains('Curr, Prin')]
    

Compute Sample Sizes

Compute the sample sizes of MDD patients and primary MDD patients who completed:

Pre-treatment assessments of different outcome domains
Post-treatment assessment of different outcome domains
Global patient ratings of overall improvements
All the above three measurements

#Complete Data of Pre-treatment RDQ
rdqPreComp = rdqPre.filter(regex = 'ID1|rdqpre_|pre_')
rdqPreComp = rdqPreComp.dropna() #Drop rows containing any missing values
rdqPreComp_dx = pd.merge(rdqPreComp, dx, on = 'ID1', how = 'inner') #Merge pre-treatment RDQ data with dianogses 

rdqPreCompMDDAll = findMDDAll(rdqPreComp_dx) #Find patients with MDD in the complete smaple
rdqPreCompMDDCurrPrin = findMDDCurrPrin(rdqPreComp_dx) #Find patients with primary MDD in the complete smaple


#Complete Data of Post-treatment RDQ
rdqPostComp = rdqPost.filter(regex = 'ID1|rdqpost|post_')
rdqPostComp = rdqPostComp.dropna() #Drop rows containing any missing values
rdqPostComp_dx = pd.merge(rdqPostComp, dx, on = 'ID1', how = 'inner') #Merge post-treatment RDQ data with dianogses 

rdqPostCompMDDAll = findMDDAll(rdqPostComp_dx) #Find patients with MDD in the complete smaple
rdqPostCompMDDCurrPrin = findMDDCurrPrin(rdqPostComp_dx) #Find patients with primary MDD in the complete smaple


#Complete Data of the Global Patient Rating of Improvement
imprvComp = imprv.dropna() #Drop rows containing any missing values
imprvComp_dx = pd.merge(imprvComp, dx, on = 'ID1', how = 'inner') #Merge post-treatment RDQ data with dianogses 

imprvCompMDDAll = findMDDAll(imprvComp_dx) #Find patients with MDD in the complete smaple
imprvCompMDDCurrPrin = findMDDCurrPrin(imprvComp_dx) #Find patients with primary MDD in the complete smaple

#Complete Data of the Above Assessments
rdqImprvDxComp = rdqPreComp.merge(rdqPostComp, on = 'ID1', how = 'inner') #Inner merging pre- and post-treatment RDQ
rdqImprvDxComp  = rdqImprvDxComp.merge(imprvComp, on = 'ID1', how = 'inner') #Inner merging Improvement
rdqImprvDxComp  = rdqImprvDxComp.merge(dx, on = 'ID1', how = 'inner') #Inner merging diagnoses

rdqImprvDxCompMDDAll = findMDDAll(rdqImprvDxComp)  #Find patients with MDD in the complete smaple
rdqImprvDxCompMDDCurrPrin = findMDDCurrPrin(rdqImprvDxComp) #Find patients with primary MDD in the complete smaple

sampleSize = pd.DataFrame([[rdqPreCompMDDAll.shape[0], rdqPreCompMDDCurrPrin.shape[0]],
                           [rdqPostCompMDDAll.shape[0], rdqPostCompMDDCurrPrin.shape[0]],
                           [imprvCompMDDAll.shape[0], imprvCompMDDCurrPrin.shape[0]],
                           [rdqImprvDxCompMDDAll.shape[0], rdqImprvDxCompMDDCurrPrin.shape[0]]],
                          columns = ['MDD ALL', 'PRINCIPLE MDD'],
                          index = ['PRE RDQ', 'POST RDQ', 'IMPROVEMENT', 'ALL ABOVE'])

sampleSize

	MDD ALL	PRINCIPLE MDD
PRE RDQ	2626	1984
POST RDQ	1926	1441
IMPROVEMENT	2118	1558
ALL ABOVE	1126	844

Label Clinically Significant Improvement: 50% Improvement

for sample in ['MDDAll', 'MDDCurrPrin']:
    dfMDD = locals().get('rdqImprvDxComp'+sample)

    labels = ['dsym', 'ndsym', 'cope', 'pmh', 'fun', 'well']

    #Compute Change Scores (Post - Pre RDQ)
    for label in labels:
        dfMDD['change_'+ label] = dfMDD.filter(regex='^post_'+label).squeeze() - dfMDD.filter(regex='^pre_'+label).squeeze()

    #Reverse symptom reduction scores (high scores = better improvement)
    dfMDD.change_dsym = dfMDD.change_dsym*(-1) 
    dfMDD.change_ndsym = dfMDD.change_ndsym*(-1)


    cut = 0.5 #Set a cutoff point
    for label in labels:
        dfMDD['change_'+label+'_cut'] = dfMDD['change_'+label] >= cut
        dfMDD['change_'+label+'_cut'] = dfMDD['change_'+label+'_cut'].replace({True: 1, False: 0}).astype('category')

    dfMDD.to_csv('../Data/dfMDD' + sample + '_satisfaction_rdq_dx_demo.csv') #Save processed data for later use
    
    

Compute the Mean of the Patient Global Rating of Improvement in Each Responder Groups

for sample in ['MDDAll', 'MDDCurrPrin']:
    dfMDD = locals().get('rdqImprvDxComp'+sample)

    #Create a contingency table
    contingencyTable = pd.crosstab(dfMDD['change_dsym_cut'], dfMDD['change_'+'ndsym'+'_cut'])

    for label in labels[2:]:
        contingencyTable = pd.concat([contingencyTable, pd.crosstab(dfMDD['change_dsym_cut'], 
                                                                    dfMDD['change_'+label+'_cut'])], axis = 1)
        percentageTable = pd.concat([percentageTable, pd.crosstab(dfMDD['change_dsym_cut'], 
                                                                    dfMDD['change_'+label+'_cut'],
                                                                  normalize='all')], axis = 1)
    #Rearrange contingency table
    rearrangeCont = {}
    for i in range(len(labels)-1):
        cont = [contingencyTable.iloc[1,2*i+1],
               contingencyTable.iloc[1,2*i],
               contingencyTable.iloc[0, 2*i+1],
               contingencyTable.iloc[0,2*i]]
        rearrangeCont[labels[i+1]] = cont

    
    rearrangeCont = pd.DataFrame(rearrangeCont)
    locals()['rearrangeCont'+sample] = rearrangeCont

    #Create a mean table
    meanTable = {}
    for label in labels[1:]:
        mean = [dfMDD[(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 1)].imprv1.mean(),
                dfMDD[(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 0)].imprv1.mean(),
                dfMDD[(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 1)].imprv1.mean(),
                dfMDD[(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 0)].imprv1.mean()]
        meanTable[label] = mean
    meanTable = pd.DataFrame(meanTable)
    meanTable

    #Create a mean improvement table with sample size in parentheses
    meanNTable = meanTable.copy()
    for i in range(meanNTable.shape[0]):
        for j in range(meanNTable.shape[1]):
            meanNTable.iloc[i,j] = str(round(meanTable.iloc[i,j],1)) + ' (' + str(rearrangeCont.iloc[i,j]) +')'

    meanNTable = meanNTable.rename(index = {0:'both', 1:'dep', 2:'other', 3:'none'})
    print(sample)
    display(meanNTable)
    meanNTable.to_csv('../Results/'+sample+'_meanImprvNTable.csv') 

MDDAll

	ndsym	cope	pmh	fun	well
both	3.2 (475)	3.2 (438)	3.3 (491)	3.3 (416)	3.3 (510)
dep	3.0 (159)	3.0 (196)	2.8 (143)	2.9 (218)	2.8 (124)
other	2.5 (102)	2.7 (159)	2.8 (142)	2.8 (119)	2.9 (176)
none	2.3 (390)	2.1 (333)	2.1 (350)	2.2 (373)	2.0 (316)

MDDCurrPrin

	ndsym	cope	pmh	fun	well
both	3.2 (366)	3.2 (335)	3.3 (375)	3.3 (317)	3.3 (393)
dep	2.9 (117)	3.0 (148)	2.7 (108)	2.9 (166)	2.7 (90)
other	2.4 (71)	2.7 (108)	2.8 (99)	2.8 (89)	2.9 (126)
none	2.2 (290)	2.1 (253)	2.1 (262)	2.1 (272)	2.0 (235)

Compute the Percentage of Higher Global Patient Ratings of Improvement in Each Responder Group

for sample in ['MDDAll', 'MDDCurrPrin']:
    dfMDD = locals().get('rdqImprvDxComp'+sample)
    rearrangeCont = locals()['rearrangeCont'+sample]
    
    #Dichotimized Improvement scores (3,4 = higher improvement) 
    #Compute the number of people with higher improvement socres in each responder group
    
    imprvN34Table = {}
    percTable = {}
    for label in labels[1:]:
        N = [sum(dfMDD[(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 1)].imprv1>=3),
                sum(dfMDD[(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 0)].imprv1>=3),
                sum(dfMDD[(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 1)].imprv1>=3),
                sum(dfMDD[(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 0)].imprv1>=3)]

        imprvN34Table[label] = N

    imprvN34Table = pd.DataFrame(imprvN34Table)
    
    #Compute the percentage of people with higher improvement scores in each responder group
    #Number of people with improvement ratings of 3 or 4 in each group/Number of responders in each group
    
    imprvPercNTable = imprvN34Table.copy()

    for i in range(imprvPercNTable.shape[0]):
        for j in range(imprvPercNTable.shape[1]):
            imprvPercNTable.iloc[i,j] = str(round(imprvN34Table.iloc[i,j]/rearrangeCont.iloc[i,j]*100,1)) + '% (' + str(rearrangeCont.iloc[i,j]) + ')'
    
    imprvPercNTable  = imprvPercNTable.rename(index = {0:'both', 1:'dep', 2:'other', 3:'none'})
    imprvPercNTable .to_csv('../Results/'+sample+'_imprvPercNTable.csv') 
    
    print(sample)
    display(imprvPercNTable)    

MDDAll

	ndsym	cope	pmh	fun	well
both	86.3% (475)	86.3% (438)	89.4% (491)	87.3% (416)	87.5% (510)
dep	76.1% (159)	78.1% (196)	64.3% (143)	77.1% (218)	68.5% (124)
other	54.9% (102)	67.3% (159)	70.4% (142)	68.1% (119)	74.4% (176)
none	45.4% (390)	37.8% (333)	38.0% (350)	40.8% (373)	32.3% (316)

MDDCurrPrin

	ndsym	cope	pmh	fun	well
both	85.8% (366)	86.0% (335)	89.1% (375)	87.1% (317)	86.8% (393)
dep	76.1% (117)	77.7% (148)	63.9% (108)	76.5% (166)	68.9% (90)
other	50.7% (71)	65.7% (108)	69.7% (99)	67.4% (89)	76.2% (126)
none	44.8% (290)	37.5% (253)	37.0% (262)	39.0% (272)	29.8% (235)

Pairwise Comparisons between Each Responder Group

We can either first do an ANOVA followed by post-hoc pairwise comparisons for those with significant ANOVA results or directly perform pairwise comparisons. Given the large sample size in the current case, I would predict all ANOVA results to be significant. I would usually skip ANOVA in this case. Here, I am only putting down ANOVA code for demonstration purposes.

#ANOVA
for sample in ['MDDAll', 'MDDCurrPrin']:
    dfMDD = locals().get('rdqImprvDxComp'+sample)

    #Create responder group labels
    for label in labels[1:]:
        dfMDD['respGroup_'+label] = 0
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 1)] = 'both'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 0)] = 'dep'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 1)] = 'other'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 0)] = 'none'

    anovaTable = pd.DataFrame()
    for label in labels[1:]:
        model = ols('imprv1 ~ C(respGroup_'+label+')', data=dfMDD).fit()
        anova = sm.stats.anova_lm(model, type = 2)
        anova = anova[['sum_sq','df','mean_sq','F','PR(>F)']] 
        anovaTable = pd.concat([anovaTable, anova])
    print(sample)
    display(anovaTable)

MDDAll

	sum_sq	df	mean_sq	F	PR(>F)
C(respGroup_ndsym)	213.646289	3.0	71.215430	84.645732	2.260267e-49
Residual	943.978045	1122.0	0.841335	NaN	NaN
C(respGroup_cope)	245.064480	3.0	81.688160	100.436278	1.369894e-57
Residual	912.559854	1122.0	0.813333	NaN	NaN
C(respGroup_pmh)	280.279624	3.0	93.426541	119.479354	3.780155e-67
Residual	877.344710	1122.0	0.781947	NaN	NaN
C(respGroup_fun)	251.285972	3.0	83.761991	103.693011	2.987838e-59
Residual	906.338362	1122.0	0.807788	NaN	NaN
C(respGroup_well)	302.513540	3.0	100.837847	132.310415	2.187827e-73
Residual	855.110793	1122.0	0.762131	NaN	NaN

MDDCurrPrin

	sum_sq	df	mean_sq	F	PR(>F)
C(respGroup_ndsym)	170.474255	3.0	56.824752	67.256812	5.535918e-39
Residual	709.709395	840.0	0.844892	NaN	NaN
C(respGroup_cope)	192.973675	3.0	64.324558	78.626084	7.826022e-45
Residual	687.209974	840.0	0.818107	NaN	NaN
C(respGroup_pmh)	223.584816	3.0	74.528272	95.345507	4.108362e-53
Residual	656.598833	840.0	0.781665	NaN	NaN
C(respGroup_fun)	208.028897	3.0	69.342966	86.658751	7.399623e-49
Residual	672.154752	840.0	0.800184	NaN	NaN
C(respGroup_well)	248.148565	3.0	82.716188	109.933135	4.801290e-60
Residual	632.035084	840.0	0.752423	NaN	NaN

As expected, we see significant group differences in global patient improvement ratings in each outcome domain.

Next, we perform pairwise comparisons based on t-tests. I compute the p-values of t-tests and cohen’s d for effect size, which is independent of sample size.

#Fucntion to compute Cohen's d for effect size
def cohend(d1, d2):
    n1, n2 = len(d1), len(d2)
    s1, s2 = np.var(d1, ddof=1), np.var(d2, ddof=1)
    s = np.sqrt(((n1 - 1) * s1 + (n2 - 1) * s2) / (n1 + n2 - 2))
    u1, u2 = np.mean(d1), np.mean(d2)
    return (u1 - u2) / s

for sample in ['MDDAll', 'MDDCurrPrin']:
    dfMDD = locals().get('rdqImprvDxComp'+sample)

    #Create responder group labels
    for label in labels[1:]:
        dfMDD['respGroup_'+label] = 0
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 1)] = 'both'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 1) & (dfMDD['change_'+label+'_cut'] == 0)] = 'dep'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 1)] = 'other'
        dfMDD['respGroup_'+label][(dfMDD['change_dsym_cut'] == 0) & (dfMDD['change_'+label+'_cut'] == 0)] = 'none'

    dfMDD['imprv_cut'] = pd.cut(dfMDD['imprv1'], [-98,2,98], labels = [0,1]) #Dichotimized improvement scores
    dfMDD['imprv_cut'] = dfMDD['imprv_cut'].astype('float') 

    respGroup = ['both','dep','other','none'] 
    sigTable = {}
    sigTableCut= {}

    for label in labels[1:]:
        sig = ['N']*6
        k = 0
        sig_cut = ['N']*6
        rowname = []
        for i in range(len(respGroup)):
            for j in range(i+1, len(respGroup)):   
                
                #t-test for the mean of improvement scores between responder groups
                res = ttest_ind(dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[i], 'imprv1'],
                         dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[j], 'imprv1']) 
                #cohen's d for the mean of improvement scores between responder groups
                d = cohend(dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[i], 'imprv1'],
                         dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[j], 'imprv1']) 

                #t-test for the dichotimized improvement scores between responder groups
                res_c = ttest_ind(dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[i], 'imprv_cut'],
                         dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[j], 'imprv_cut'])
                #cohen's d for the dichotimized improvement scores between responder groups
                d_c = cohend(dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[i], 'imprv_cut'],
                         dfMDD.loc[dfMDD['respGroup_'+label] == respGroup[j], 'imprv_cut'])


                #Record p-values
                sig[k] = 'p=' +str(round(res[1],3)) + '; d='+str(round(d,2))
                sig_cut[k] = 'p=' +str(round(res_c[1],3)) + '; d='+str(round(d_c,2))

                k+=1

                rowname.append(respGroup[i]+ '-' +respGroup[j])

        sigTable[label] = sig
        sigTableCut[label] = sig_cut


    renameIndex = {}
    for i in range(len(rowname)):
            renameIndex[i] = rowname[i]

    sigTable = pd.DataFrame(sigTable)
    sigTable = sigTable.rename(index = renameIndex)
    sigTable.to_csv('../Results/'+sample+'meanImprvPairWiseComp.csv')

    sigTableCut = pd.DataFrame(sigTableCut).rename(index = renameIndex)
    sigTableCut.to_csv('../Results/'+sample+'percImprvPairWiseComp.csv')
    
    print(sample,': Continuous Improvement Scores')
    display(sigTable)
    print(sample,': Dichotimized Improvement Scores')
    display(sigTableCut)

MDDAll : Continuous Improvement Scores

	ndsym	cope	pmh	fun	well
both-dep	p=0.0; d=0.37	p=0.004; d=0.25	p=0.0; d=0.71	p=0.0; d=0.47	p=0.0; d=0.64
both-other	p=0.0; d=0.89	p=0.0; d=0.59	p=0.0; d=0.6	p=0.0; d=0.62	p=0.0; d=0.52
both-none	p=0.0; d=1.05	p=0.0; d=1.22	p=0.0; d=1.33	p=0.0; d=1.2	p=0.0; d=1.42
dep-other	p=0.0; d=0.47	p=0.002; d=0.33	p=0.518; d=-0.08	p=0.128; d=0.17	p=0.394; d=-0.1
dep-none	p=0.0; d=0.67	p=0.0; d=0.93	p=0.0; d=0.63	p=0.0; d=0.78	p=0.0; d=0.75
other-none	p=0.026; d=0.25	p=0.0; d=0.6	p=0.0; d=0.68	p=0.0; d=0.59	p=0.0; d=0.84

MDDAll : Dichotimized Improvement Scores

	ndsym	cope	pmh	fun	well
both-dep	p=0.002; d=0.28	p=0.009; d=0.22	p=0.0; d=0.71	p=0.001; d=0.28	p=0.0; d=0.52
both-other	p=0.0; d=0.84	p=0.0; d=0.5	p=0.0; d=0.55	p=0.0; d=0.52	p=0.0; d=0.36
both-none	p=0.0; d=0.97	p=0.0; d=1.18	p=0.0; d=1.31	p=0.0; d=1.12	p=0.0; d=1.42
dep-other	p=0.0; d=0.46	p=0.023; d=0.24	p=0.275; d=-0.13	p=0.073; d=0.21	p=0.265; d=-0.13
dep-none	p=0.0; d=0.64	p=0.0; d=0.87	p=0.0; d=0.54	p=0.0; d=0.78	p=0.0; d=0.78
other-none	p=0.087; d=0.19	p=0.0; d=0.61	p=0.0; d=0.68	p=0.0; d=0.56	p=0.0; d=0.92

MDDCurrPrin : Continuous Improvement Scores

	ndsym	cope	pmh	fun	well
both-dep	p=0.0; d=0.4	p=0.002; d=0.31	p=0.0; d=0.78	p=0.0; d=0.54	p=0.0; d=0.67
both-other	p=0.0; d=1.0	p=0.0; d=0.65	p=0.0; d=0.64	p=0.0; d=0.64	p=0.0; d=0.48
both-none	p=0.0; d=1.08	p=0.0; d=1.24	p=0.0; d=1.36	p=0.0; d=1.28	p=0.0; d=1.48
dep-other	p=0.001; d=0.53	p=0.01; d=0.33	p=0.498; d=-0.09	p=0.332; d=0.13	p=0.245; d=-0.16
dep-none	p=0.0; d=0.66	p=0.0; d=0.9	p=0.0; d=0.61	p=0.0; d=0.8	p=0.0; d=0.79
other-none	p=0.18; d=0.18	p=0.0; d=0.58	p=0.0; d=0.67	p=0.0; d=0.64	p=0.0; d=0.93

MDDCurrPrin : Dichotimized Improvement Scores

	ndsym	cope	pmh	fun	well
both-dep	p=0.014; d=0.26	p=0.024; d=0.22	p=0.0; d=0.7	p=0.003; d=0.29	p=0.0; d=0.49
both-other	p=0.0; d=0.93	p=0.0; d=0.53	p=0.0; d=0.56	p=0.0; d=0.53	p=0.005; d=0.29
both-none	p=0.0; d=0.97	p=0.0; d=1.17	p=0.0; d=1.33	p=0.0; d=1.16	p=0.0; d=1.47
dep-other	p=0.0; d=0.55	p=0.034; d=0.27	p=0.378; d=-0.12	p=0.119; d=0.21	p=0.235; d=-0.16
dep-none	p=0.0; d=0.65	p=0.0; d=0.87	p=0.0; d=0.56	p=0.0; d=0.81	p=0.0; d=0.85
other-none	p=0.375; d=0.12	p=0.0; d=0.58	p=0.0; d=0.68	p=0.0; d=0.59	p=0.0; d=1.04

Overall, you can see that patients with 50% improvement in depressive symptom reduction and other outcome domains reported significantly higher global ratings of improvement than patients with improvements in only one outcome domain. Patients with 50% improvement in either depressive symptom reduction or other outcome domains reported higher global improvement ratings than nonresponders. Lastly, there was no significant difference in global improvement ratings between patients who only responded to depressive symptom reduction or other treatment outcome domains.

Such findings indicate that the improvement in symptom reduction is no more critical than other treatment outcome domains. We should consider adding other treatment outcome domains to standard outcome metrics to capture the full spectrum of depression remission.

Share: Twitter, Facebook