A Python module of MetaAnalysis, usually applied in systemtic reviews of Evidencebased Medicine.
Project description
PythonMeta
Info
Name = PythonMeta
Version = 1.26
Author = 邓宏勇Deng Hongyong
Email = dephew@126.com
URL = www.pymeta.com
Date = 2021.11.17
History
2019.7.23 Ver 1.11 released.
2021.1.20 Ver 1.20 released. Fixed a bug while calculating random effect subgroup's weight.
2021.1.22 Ver 1.23 released. Fixed a bug of Funnel plot.
2021.10.16 Ver 1.25 released. 1, Changes for funnel plot; 2, Egger's test for publication bias.
2021.11.17 Ver 1.26 released. Some minor changes.
About
This is a MetaAnalysis package.
This module was designed to perform some Evidencebased medicine (EBM) tasks, such as:
 Combining effect measures (OR, RR, RD for count data and MD, SMD for continuous data);
 Heterogeneity test(Q/Chisquare test);
 Subgroup analysis;
 Plots drawing: forest plot, funnel plot, etc.
Statistical algorithms in this software cited from: Jonathan J Deeks and Julian PT Higgins, on behalf of the Statistical Methods Group of The Cochrane Collaboration. Statistical algorithms in Review Manager 5, August 2010.
Please cite me in any publictions like: Deng Hongyong. PythonMeta, Python module of Metaanalysis, cited 20xxxxxx (or your time); 1 screen(s). Available from URL: http://www.pymeta.com
This is an ongoing project, so, any questions and suggestions from you are very welcome.
Installing
Install and update using pip
:
pip install PythonMeta
Import
Import the PythonMeta module in your code:
import PythonMeta
Functions and Classes
There are four functions/classes in PythonMeta package:
Help()(function): Show help information of PythonMeta.
Data()(class): Set and Load data to analysis.
 datatype (attribute, string): set the data type:'CATE' for CATEgorical/count/binary/dichotomous data, or 'CONT' for continuous data.
 studies (attribute, array): Array to store the study data.
 subgroup (attribute, array): Array to store the subgroup.
 nototal (attribute, binary): Flag of do NOT calculate the total effect.
 readfile(filename) (method): Read data file.
 Input: filename(string) (e.g. "c:\1.txt");
 Output: lines array (always as input of method getdata()). (See Sample code and data files)
 getdata(lines) (method): Load data into attribute array of 'studies'.
 Input: lines array (always from method readfile());
 Output: attribute 'studies'. (See Sample code and data files)
Meta()(class): Set and perform the MetaAnalysis.
 datatype (attribute, string): set the data type:'CATE' for CATEgorical/count/binary/dichotomous data, or 'CONT' for CONTinuous data. Attention: this attribute should same to Data().datatype.
 studies (attribute, array): Array of study data to metaanalysis.
 subgroup (attribute, array): Array to store the subgroup. Attention: this attribute should same to Data().subgroup.
 nototal (attribute, binary): Flag of do NOT calculate the total effect. Attention: this attribute should same to Data().nototal.
 models (attribute, string): set effect models as 'Fixed' or 'Random'.
 effect (attribute, string): set effect size as 'OR':odds ratio; 'RR': risk ratio; 'RD':risk difference; 'MD':weighted mean diff; 'SMD':standard mean diff.
 algorithm (attribute, string): set the algorithms of metaanalysis: 'MH':MantelHaenszel;'Peto';'IV':Inverse variance;'IVHeg'(DEFAULT),'IVCnd','IVGls':for SMD algorithms
 meta(studies, nosubgrp=False) (method): perform the metaanalysis.
 Input: 1, studies array (always from Data().getdata); 2, nosubgrp flag, False as default.
 Output: result array [[Total...],[study1...],[subgroup1,...],[studyn,...]...[subgroupk,...]]. (See Sample code for more information)
 Eggers_test (metarults) (method): Egger's test for publication bias.
 Input: result from self.meta();
 Output: (Intercept(far away from 0 means bias), tvalue, pvalue(<0.05 means bias), std. dev., CI).
Fig()(class): Set and draw the result figures.
 size (attribute, integer array): set the canvas size in inchs, default [6,6].
 dpi (attribute, integer): set the resolution of figure (dot per inch), default 80pts.
 title (attribute, string): set the title of figure.
 nototal (attribute, binary): Flag of do NOT show the total effect, default False.
 forest(results) (method): drawing the forest plot.
 Input: results array, always from Meta().meta.
 Output: matplotlib.pyplot.figure object; (See Sample code for more information)
 funnel(results) (method): drawing the funnel plot.
 Input: results array, always from Meta().meta;
 Output: matplotlib.pyplot.figure object. (See Sample code for more information)
Example
Sample code: sample.py
import PythonMeta as PMA
def showstudies(studies,dtype):
#show continuous data
if dtype.upper()=="CONT":
text = "%10s %30s %30s \n"%("Study ID","Experiment Group","Control Group")
text += "%10s %10s %10s %10s %10s %10s %10s \n"%(" ","m1","sd1","n1","m2","sd2","n2")
for i in range(len(studies)):
text += "%10s %10s %10s %10s %10s %10s %10s \n"%(
studies[i][6], #study ID
str(studies[i][0]), #mean of group1
str(studies[i][1]), #SD of group1
str(studies[i][2]), #total num of group1
str(studies[i][3]), #mean of group2
str(studies[i][4]), #SD of group2
str(studies[i][5]) #total num of group2
)
return text
#show dichotomous data
text = "%10s %20s %20s \n"%("Study ID","Experiment Group","Control Group")
text += "%10s %10s %10s %10s %10s \n"%(" ","e1","n1","e2","n2")
for i in range(len(studies)):
text += "%10s %10s %10s %10s %10s \n"%(
studies[i][4], #study ID
str(studies[i][0]), #event num of group1
str(studies[i][1]), #total num of group1
str(studies[i][2]), #event num of group2
str(studies[i][3]) #total num of group2
)
return text
def showresults(rults):
text = "%10s %6s %18s %10s"%("Study ID","n","ES[95% CI]","Weight(%)\n")
for i in range(1,len(rults)):
text += "%10s %6d %4.2f[%.2f %.2f] %6.2f\n"%( # for each study
rults[i][0], #study ID
rults[i][5], #total num
rults[i][1], #effect size
rults[i][3], #lower of CI
rults[i][4], #higher of CI
100*(rults[i][2]/rults[0][2]) #weight
)
text += "%10s %6d %4.2f[%.2f %.2f] %6d\n"%( # for total effect
rults[0][0], #total effect size name
rults[0][5], #total N (all studies)
rults[0][1], #total effect size
rults[0][3], #total lower CI
rults[0][4], #total higher CI
100
)
text += "%d studies included (N=%d)\n"%(len(rults)1,rults[0][5])
text += "Heterogeneity: Tau\u00b2=%.3f "%(rults[0][12]) if not rults[0][12]==None else "Heterogeneity: "
text += "Q(Chisquare)=%.2f(p=%s); I\u00b2=%s\n"%(
rults[0][7], #Q test value
rults[0][8], #p value for Q test
str(round(rults[0][9],2))+"%") #Isquare value
text += "Overall effect test: z=%.2f, p=%s\n"%(rults[0][10],rults[0][11]) #ztest value and pvalue
return text
def main(stys,settings):
d = PMA.Data() #Load Data class
m = PMA.Meta() #Load Meta class
f = PMA.Fig() #Load Fig class
#You should always tell the datatype first!!!
d.datatype = settings["datatype"] #set data type, 'CATE' for binary data or 'CONT' for continuous data
studies = d.getdata(stys) #load data
#studies = d.getdata(d.readfile("studies.txt")) #get data from a data file, see examples of data files
print(showstudies(studies,d.datatype)) #show studies
m.subgroup=d.subgroup #set the subgroup
m.datatype=d.datatype #set data type for metaanalysis calculating
m.models = settings["models"] #set effect models: 'Fixed' or 'Random'
m.algorithm = settings["algorithm"] #set algorithm, based on datatype and effect size
m.effect = settings["effect"] #set effect size:RR/OR/RD for binary data; SMD/MD for continuous data
results = m.meta(studies) #performing the analysis
print(m.models + " " + m.algorithm + " " + m.effect)
print (showresults(results)) #show results table
f.forest(results).show() #show forest plot
f.funnel(results).show() #show funnel plot
print(m.Eggers_test(results)) #Eggers_test: (Intercept, tvalue, pvalue, SD and CI)
if __name__ == '__main__':
samp_cate=[ #this array can be stored into a data file by lines, and loaded with d.readfile("filename")
"Fang 2015,15,40,24,37",
"Gong 2012,10,40,18,35",
"Liu 2015,30,50,40,50",
"Long 2012,19,40,26,40",
"Wang 2003,7,86,15,86",
"<subgroup>name=short term",
"Chen 2008,20,60,28,60",
"Guo 2014,31,51,41,51",
"Li 2015,29,61,31,60",
"Yang 2006,21,40,31,40",
"Zhao 2012,27,40,30,40",
"<subgroup>name=medium term",
"#<nototal>",
" ",
"#This is a sample of binary data with subgroup.",
"#Syntax: study name, e1, n1, e2, n2",
"#e1,n1: events and number of experiment group;",
"#e2,n2: events and number of control group.",
"#And you can add a line of <nototal> to hide the Overall result."]
samp_cont=[ #this array can be stored into a data file by lines, and loaded with d.readfile("filename")
"Atmaca 2005, 20.9, 6.0, 15, 27.4, 8.5, 14",
"Guo 2014, 12.8, 5.2, 51, 11.9, 5.3, 51",
"Liu 2010, 23.38, 5.86, 35, 24.32, 5.43, 35",
"Wang 2012, 15.67, 8.78, 43, 18.67, 9.87, 43",
"Xu 2002, 15.49, 7.16, 50, 21.72, 8.07, 50",
"Zhao 2012, 12.8, 5.7, 40, 13.0, 5.2, 40",
" ",
"#This is a sample of continuous data.",
"#Input one study in a line;",
"#Syntax: study name, m1, sd1, n1, m2, sd2, n2",
"#m1, sd1, n1: mean, SD and number of experiment group;",
"#m2, sd2, n2: mean, SD and number of control group."]
#sample 1: dichotomous data
settings={
"datatype":"CATE", #for CATEgorical/count/binary/dichotomous data
"models":"Fixed", #models: Fixed or Random
"algorithm":"MH", #algorithm: MH, Peto or IV
"effect":"RR"} #effect size: RR, OR, RD
main(samp_cate,settings)
#sample 2: continuous data
settings={
"datatype":"CONT", #for CONTinuous data
"models":"Fixed", #models: Fixed or Random
"algorithm":"IV", #algorithm: IV
"effect":"MD"} #effect size: MD, SMD
main(samp_cont,settings)
Or you can load data from a file, like:
studies = d.getdata(d.readfile("studies.txt")
Here are some examples of data file: (Please remember all lines start with # are comment lines, which will be ignored while loading.)
Sample of continuous data
Atmaca 2005, 20.9, 6.0, 15, 27.4, 8.5, 14
Guo 2014, 12.8, 5.2, 51, 11.9, 5.3, 51
Liu 2010, 23.38, 5.86, 35, 24.32, 5.43, 35
Wang 2012, 15.67, 8.78, 43, 18.67, 9.87, 43
Xu 2002, 15.49, 7.16, 50, 21.72, 8.07, 50
Zhao 2012, 12.8, 5.7, 40, 13.0, 5.2, 40
#This is a sample of continuous data.
#Input one study in a line;
#Syntax: study name, m1, sd1, n1, m2, sd2, n2
#m1, sd1, n1: mean, SD and number of experiment group;
#m2, sd2, n2: mean, SD and number of control group.
Sample of dichotomous data
Fang 2015, 15, 40, 24, 37
Gong 2012, 10, 40, 18, 35
Liu 2015, 30, 50, 40, 50
Long 2012, 19, 40, 26, 40
Pan 2015a, 57, 100, 68, 100
Wang 2001, 13, 18, 17, 18
Wang 2003, 7, 86, 15, 86
#This is a sample of binary data.
#Input one study in a line;
#Syntax: study name, e1, n1, e2, n2
#e1,n1: events and number of experiment group;
#e2,n2: events and number of control group.
Sample of data with subgroup
Fang 2015,15,40,24,37
Gong 2012,10,40,18,35
Liu 2015,30,50,40,50
Long 2012,19,40,26,40
Wang 2003,7,86,15,86
<subgroup>name=short term
Chen 2008,20,60,28,60
Guo 2014,31,51,41,51
Li 2015,29,61,31,60
Yang 2006,21,40,31,40
Zhao 2012,27,40,30,40
<subgroup>name=medium term
#<nototal>
#This is a sample of subgroup.
#Cumulative metaanalysis and Senstivity analysis will blind to all <subgroup> tags.
#And you can add a line of <nototal> to hide the Overall result.
Please download all above sample code and data files (maybe updated) at www.pymeta.com.
Contact
Deng Hongyong Ph.D
Shanghai University of Traditional Chinese Medicine
Shanghai, China 201203
Email: dephew@126.com
Web: www.PyMeta.com
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for PythonMeta1.26py3noneany.whl
Algorithm  Hash digest  

SHA256  2826030347c284c0d2a0b8c81b132d45567146265ec0205bb8f4628fcab5e6e4 

MD5  db571e2f2f885af9775024c522623e6e 

BLAKE2b256  bb38cdcde9f02cd41595d74c6a32e81f2f20a37d4e045be7ab901dcbb1b39751 