Canibais e Reis

File: china-project.xlsm (27.4 Mb)

"The China-Oxford-Cornell Study on Dietary, Lifestyle and Disease Mortality Characteristics in 65 Rural Chinese Counties was a study comparing the diets, lifestyle and disease characteristics of populations of 65 rural counties in China in the 1970s and 1980s. The study only compared the prevalence of disease characteristics. It did not evaluate all causes of death, such as accidents. Professor T. Colin Campbell of Cornell, led the first two major studies in the 1980s and 1990s. In 1991, The New York Times called the China-Oxford-Cornell study ‘the Grand Prix of epidemiology’. Campbell’s summary of the results of this and other studies appeared in his 2005 book The China Study. The findings of the extensive research in the study pointed that some diseases of affluence were caused by Westernisation, especially the growing consumption of animal protein and dairy products, previously not largely common or unknown in China. The study was jointly funded by the Universities of Oxford, Cornell and the Government of China."

 

Source: Wikipedia.

CHINA PROJECT, in friendly Excel format

Following to the work done by independent researcher Denise Minger (at Raw Food SOS), analysing some epidemiological data from the CHINA STUDY, an analysis which is generating a quite interesting on-line debate (see all updated links in the end of this post), I just noticed the full the data from the so called CHINA PROJECT is available on-line, from this page: Geographic study of mortality, biochemistry, diet and lifestyle in rural China. Unfortunately, this oficial data is not provided in a user friendly format, so I decided to convert it to Excel format, and provide it to other (part-time?) researchers like me, so they can also do their own analysis. This large 27.4 Mb Excel file includes all data from the CHINA PROJECT surveys of 1983, 1989 and 1993 (mainland data from oficial files CH83M.CSV, CH83PRU.CSV, CH83DG.CSV, CH83Q.CSV, CH89M.CSV, CH89PRU.CSV, CH89DG.CSV, CH89Q.CSV, CH93PRU.CSV and CH93Q.CSV; first mainland mortality survey refers to 1975, so file CH83M.CSV should be renamed to CH75M.CSV; also the Taiwan data, in which the survey data refers to 1989, and the mortality data refers to 1986-88; for the sake of simplicity, I considered the Taiwan mortality as if it was from 1989; Taiwan files are CHTAIM.CSV, CHTAIPRU.CSV and CHTAIQ.CSV). The final Excel database can be downloaded from here (right click below the link and select Save Link As):

File: china-project-data.xlsm (27.4 Mb)
Excel file with (almost) full CHINA PROJECT raw data. Since some workbooks have large number of rows/columns, you must use Excel 2007 or a higher version to view this file.

As you can see, this is a truly huge database, with 639 items/variables from groups M, P, D and Q, obtained in the 1983 (65 counties), 1989 (69+15 counties) and 1993 surveys, comprising biochemistry, diet and lifestyle items from China & Taiwan. It’s no wonder the New York Times once called the China-Oxford-Cornell study the Grand Prix of Epidemiology! "In the 1983 survey, 367 items of information were collected on how people live and how they die in 138 rural Chinese villages; 6500 adults and their families were surveyed. In the 1989-90 survey, more than 1000 items of information were collected in 170 villages in rural China and Taiwan, involving 10,200 adults and their families." We have here a total 972 food & health items/variables (many are repeated, because of the different surveys/years) , from a total of 85 China and Taiwan counties/regions (69 from mainland China and 15 from Taiwan). For a detailed description of the methods used in the CHINA PROJECT, please see this brief description and also this document. Also, take some time to see the oficial documents/statistics of the CHINA STUDY (correlations are included):

CHINA STUDY oficial documents:

Study description and methods
Summary (simple) statistics for all 639 variables
Statistics/correlations: Mortality, laboratory, diet and questionaire
Questionaire: Full listings of the six questionnaires
ANNEX: Age-specific deaths and death rates in urban and rural China

 

Source: Geographic study of mortality, biochemistry, diet and lifestyle in rural China.

 

CHINA STUDY files:

China Study I (1983-1984 survey)
Data for this study include blood, urine, and food samples, as well as questionnaire and 3-day diet information for a total ‘of 6500 people in 65 counties in rural China were collected in 1983-1984. Data for 1973-1975 mortality rates for nearly 50 ‘kinds of cancers and other disease were obtained.

China Study II (1989-1990 survey)
The same 65 counties and 6500 individuals from Study I were re-surveyed in 1989-1990. Twenty new counties in mainland China and Taiwan were added, resulting in a total of 10,200 participants. Mortality data for 1986-1988 were obtained as well.

The China Project Data
All data are in comma-delimited format, which means commas separate the values for each variable. Each column represents a variable. The first 3 columns/variables in every data set are:
County = the Chinese county surveyed
Sex = M (male), F (female), T (both)
Xiang = the village within each county that was surveyed: 1 (village 1), 2 (village 2), 3 (both villages)

Data files for Study I (1983-1984)
CH83M.CSV: 1973-1975 mainland mortality rates, provided for ages 0-34 and 35-69
CH83PRU.CSV: mainland plasma, red blood cell, and urine variables (eg. average cholesterol)
CH83DG.CSV: mainland data for diet and geographic variables (eg. animal protein)
CH83Q.CSV: mainland questionnaire data (eg. smoking habits)

Data files for Study II (1989-1990)
CH89M.CSV: 1986-1988 mainland mortality rates, provided for ages 0-34 and 35-69
CH89PRU.CSV: mainland data for plasma, red blood cell, and urine variables
CH89DG.CSV: mainland data for diet and geographic variables
CH89Q.CSV: mainland questionnaire data
CHTAIM.CSV: 1986-1988 Taiwan mortality rates, provided for ages 0-34 and 35-69
CHTAIPRU.CSV: Taiwan plasma, red blood cell, and urine variables
CHTAIQ.CSV: Taiwan questionnaire data

Additional Files
CH93PRU.CSV: 1993 mainland plasma, red blood cell, and urine variables
CH93Q.CSV: 1993 mainland plasma, red blood cell, and urine variables

Data dictionary
CHNAME.TXT
A data dictionary is a list of all variables, including their variable name (eg. M030=liver cancer mortality for ages 0-34) and variable description. This information is vital to an understanding of the data and what each variable name represents. Note that mortality variables start with "M", diet variables start with "D", plasma/blood/urine variables start with "P"/"R"/"U", ‘and questionnaire variables start with "Q".

Important note
Many variables are missing sex-specific and/or village-specific data. Therefore, it is recommended that all analyses are restricted to sex=T (i.e., both sexes) and xiang=3 (i.e., both villages).

 

Source: 30 Bananas a Day

 

Huge database, isn’t it? Here is a pdf file listing all the 639 items colected:

File: china-project-items.pdf (122 kb)
This is a listing of all 639 CHINA PROJECT items/variables. It has 18 pages. Legend: D-diet survey; G-general features; M-mortality; P-Plasma; Q-Questionaire; R-Red blood cell; U-Urine / nitrosamine study.

Figure: CHINA PROJECT 1989 survey areas.

 

Correlation analysis of the 1989 data

Besides "hunter-gathering" all the raw data into an Excel file (see workbook "FULL"), I also started my own analysis by finding all possible correlations on a "smaller" database (see workbook "1989"), which refers to the data only from 1989, which is significantly larger than the data from 1983. This "smaller" database has the same number of items, the 639 items (a few of them are empty) from groups M, P, D and Q, also from the 85 counties/regions, and only the record "Total xiang [=administrative area] code 3", which represents the average values for men and female and from xiangs 1 and 2. Except for certain items that are men or women specific, I used the correspoding M or F values. This smaller database is included in the Excel file, but It is also provided here in pdf format. This must be printed in A3 format or larger size, otherwise letters will be too small for reading (you can magnify it in the Adobe Acrobat viewer):

File: china-project-1989.pdf (308 kb)
This file has all 1989 data from the CHINA PROJECT, in 16 pages, listing values for 639 items from 85 counties.

For the n=639 items/variables there are n*(n-1)/2  = 203.841 (more than two hundred thousand) associations, for which I calculated Pearson correlations using Excel’s built in function CORREL (see workbook "CORR"). All these correlations, ranging all the way from -1 to +1, are included in the Excel file. If you’re interested, you can calculate any correlation between any 2 items using this cell function: =CORREL(mycol(x),mycol(y)), where x and y are the numbers of columns in workbook "1989" for the items you want to find the correlation. For example, if you want the correlation between M001(89) and M002(89), you would have to write in a cell this: =CORREL(mycol(1),mycol(2)), and the result would be 0.6504. To find the corresponding column number for item M001(89), you can use the function =MATCH("M001(89)",PARAM1989,0), which would give the result 1. Well, from those thousands of correlations, 151.568 correlations (good or null) are between items from different groups (M, P, D and Q). And guess what? 10.631 of these are highly significant (p8.000 Dr. Campbell often refers to. Actually, there are 3.697 correlations with r=>0.5 and phere. The corresponding VBA function for the P-Values is this one:

'P-Values
Function pvalue(correlation As Single, samplesize As Integer) As Single
 freedom = samplesize - 2
 zvalue = Abs(correlation) * (freedom / (1 - correlation ^ 2)) ^ 0.5
 pvalue = Application.TDist(zvalue, freedom, 2)
End Function

Correlation is a good measure of the strength of a association, in this case a linear association (you would have to use another statistic, like the Spearman Rank Association to evaluate non-linear associations). For samples of similar size, which happen here in 100% of cases, since all sample sizes are 52 and 85), the higher the correlation, the higher its significance and the lower the P-Value will be. I would say most correlations of the CHINA PROJECT above +/-0.5, because of their average sample size of 71, are usually highly significant, with p

Function mysamplesize(col1 As Integer, col2 As Integer) As Integer
 mysamplesize = 85
 For i = 1 To 85
 If Range("DATA1989").Cells(i, col1) = "" Or Range("DATA1989").Cells(i, col2) = "" Then mysamplesize = mysamplesize - 1
 Next i
End Function

Ned Kock has an interesting post on this subject, explaining that with a large enough sample, anything is significant. Actually, P-values are very sensitive to sample size. Try some pairs of correlation/sample size values on the formula above and you’ll understand this. This is why it is quite easy to find significant associations in large studies like the CHINA PROJECT even when the correlations are very modest.

I suppose this might be already a good start to find out wich items/variables might be more relevant, to then establish meaningful associations in a more indepth multivariate analysis, like the one Ned Kock, from the Health Correlator already started. Anyway, even those "fewer" correlations from different groups are too many to be analysed as a whole (don’t try to print them all, unless you don’t mind printing 2.707 A4 pages), so I wrote a realy nice Excel VBA macro to filter the whole data according to a certain criteria. These criteria include (i) limiting the items to be listed, for example show only items D038, Q158, D045 and Q078; (ii) specify a minimum absolute correlation value, use 0.5 or higher to obtain stronger results; (iii) and/or show correlations only from different groups (M, P, D and Q), in order to avoid irrelevant associations of strictly related items. Just one note: Excel 2007 is very anoying on what concerns macros, it keeps giving security warnings if you don’t disable them. To activate the macro feature in Excel 2007, or disable all warning, please refer to these instructions.

 

CHINA PROJECT full list of correlations

Here are some simple examples of how this Excel macro works. Suppose you want a full listing of valid correlations, from all 639 items, sorted by higher correlation, and only correlations from different groups. Open the Excel file "china-project-data.xlsm" with Excel 2007 or higher version, hold the Control key, and then press the key ‘m’ (CTRL+m) to call the VBA macro "myfilter". A first window will show, asking you to select yout favourite items. Since we now want all items, leave the entry box empty and simply press OK. A second window will show, asking for the minimum correlation. In order to avoid a large results list, try 0.5 or higher. In the third/last window, for the same reason, select YES for correlations only from different groups. This query shall produce a list of the 3.697 CHINA STUDY strongest correlations, 91 A4 pages, something like this. One trick: if you want to reverse to the original listing, press CTRL+a to run the macro ‘ShowAllRecords’.

File: china-project-full-corr05-different.pdf (742 kb)
CHINA PROJECT: All 639 items in the CHINA PROJECT: 3.697 strongest linear associations, items from different groups, absolute correlation above 0.5. This document has 91 pages.

Figures: Excel VBA macro. Sequence of input boxes.

If you don’t limit your queries to only correlations from different groups, you will notice, from the full list of correlations (or linear associations) that, as expected, many of the strongest of them are from pairs of items from the same group (M, P, D or Q) This happens because quite a few of them are strictly related, like "D002-Total lipid intake/gr/day)" and "D005-% of caloric intake from fat". In order to find more useful associations, I suggest seeking only for items from different groups. More interesting examples below, keep reading.

 

CHINA PROJECT fat related correlations

Now let’s try to get to some more interesing associations. For example, lets investigate all things related to fat intake. We should first take a look at the items list (workbook "ITEMS") and see which ones are best related to fat intake and blood lipids. So, from the dozens of fat related items in the CHINA PROJECT, we could choose , for example, all these:

D002 TOTFAT diet survey TOTAL LIPID INTAKE (g/day/reference man)
D053 ANIMFAT diet survey ADDED ANIMAL FAT (for cooking, spreading etc) INTAKE (g/day/reference man)
D055 ADDEDFAT diet survey TOTAL ADDED FAT (5144 + 5145) INTAKE (g/day/reference man)

D082 MUFA diet survey MONOUNSATURATED FATTY ACID INTAKE (g/day/reference man)
D083 PUFA diet survey POLYUNSTURATED FATTY ACID INTAKE (g/day/reference man)
D084 SATFA diet survey SATURATED FATTY ACID INTAKE (g/day/reference man)

D092 TOTn3 diet survey TOTAL n3 POLYUNSATURATED FATTY ACID INTAKE (g/day/reference man)
D093 TOTn6 diet survey TOTAL n6 POLYUNSATURATED FATTY ACID INTAKE (g/day/reference man)
D094 TOTn9 diet survey TOTAL n9 MONOUNSATURATED FATTY ACID INTAKE (g/day/reference man)

Q164 dOILFAT questionnaire DAILY CONSUMPTION OF OIL AND FAT (g/day)
Q168 dANIMFAT questionnaire DAILY CONSUMPTION OF ANIMAL FAT (g/day)
Q169 dVEGFAT questionnaire DAILY CONSUMPTION OF VEGETABLE FAT (g/day)

R003 SATFA red blood cell TOTAL LIPID SATURATES (14:0+16:0+18:0+20:0+22:0+24:0) (% of total fatty acid by weight)
R004 MUFA red blood cell TOTAL LIPID MONOUNSATURATES (16:1+18:1+20:1+22:1+24:1) (% of total fatty acid by weight)
R005 TOTn6 red blood cell TOTAL LIPID n6 POLYUNSATURATES (18:2+20:2+20:3+20:4) (% of total fatty acid by weight)
R006 TOTn3 red blood cell TOTAL LIPID n3 POLYUNSATURATES (18:3+20:5+22:6) (% of total fatty acid by weight)

So how do we query our Excel database about these oily items? Easy as the example above, but anyway let’s see step by step how it’s done. The variables we want to search for are these:

D002,D053,D055,D082,D083,D084,D092,D093,D094,Q164,Q168,
Q169,R003,R004,R005,R006

Let’s run the VBA macro again. Simply press CRL+m and the first window will ask for the items. Copy/paste the line above into the input box, and press OK. Then select a minimum correlation value of, for example, 0.5 and press OK. Let’s try correlations of items from different groups, so in the third/last window press YES. The final result is a list of 286 associations, in 7 pages, that will look like this. I suggest you check the values of these correlations against the oficial diet correlations. Notice the correlations in the CHINA PROJECT were calculate as percentages. To convert them to the real values, simply divide by 100. For example, the oficial correlation of items D001 and D002 is on page 482 of the diet correlations document and it’s value is +27, or r=0.27. If you go to workbook "1989" of the Excel file, you will notice the columns for D001(89) and D002(89) are nr. 227 and 228. So the Excel correlation would be calculated from the formula =CORREL(mycol(227),mycol(228)), which gives a result of 0.271. This is exactly the same value, and trust me, I dind’t cherry pick these items just to obtain a "convenient" result. Try other items yourself, and notice most results will be very close, if not equal, to the oficial correlations (except for mortality items, see explanation in the end of this article).

File: china-project-fats-corr05-different.pdf (135 kb)
CHINA PROJECT: Lipid/fats related items: 286 associations, items from different groups, absolute correlation above 0,5.

 

CHINA PROJECT carbs/sugar and fruit

Let’s try another example, this one regarding carbohydrate, wheat, flour, fruit and sugar intake. I know Dr. William Davis would like to see these associations, so let’s look again at the CHINA PROJECT items list and select relevant items/variables. I would select these ones:

D004 SOLCARB diet survey CARBOHYDRATE (nitrogen free extract) INTAKE (g/day/reference man)
D009 %CARBKCAL diet survey PERCENTAGE OF CALORIC INTAKE FROM CARBOHYDRATE (for reference man)

Q158 dWHEAT questionnaire DAILY CONSUMPTION OF WHEAT (g/day air-dry basis)
D037 RICE diet survey RICE INTAKE (g/day/reference man, air-dry basis)
D038 WHTFLOUR diet survey WHEAT FLOUR INTAKE (g/day/reference man, air-dry basis)
D039 OTHCEREAL diet survey OTHER CEREAL INTAKE (g/day/reference man, air-dry basis)
D040 STCHTUBER diet survey STARCHY TUBER INTAKE (g/day/reference man, fresh weight)

D045 FRUIT diet survey FRUIT INTAKE (g/day/reference man, fresh weight)
Q173 dFRUIT questionnaire DAYS PER YEAR EAT FRUIT

D056 STCHSUGAR diet survey PROCESSED STARCH AND SUGAR INTAKE (g/day/reference man, as-consumed basis)

So, again, we already know which are our input items. Here they are:

D004,D009,Q158,D037,D038,D039,D040,D045,Q173,D056

Now press CRTL+m to run the Excel macro, and copy/paste the line above into the input box. Press OK. I’ll choose again a minimum correlation of 0.5 and associations of items from different groups only. The final result is 178 associations that can be printed in 5 A4 pages. The corresponding file is this one. Remember: if you want to reverse to the original full items list, press CTRL+a to run the macro ‘ShowAllRecords’.

File: china-project-carbs-corr05-different.pdf (111 kb)
CHINA PROJECT: Carbs, wheat, flour, fruit and sugar items: 178 associations, items from different groups, absolute correlation above 0,5.

 

CHINA PROJECT heart disease mortality

Now another example, all things heart related. I mean cardiovascular mortality. The related items we have from the items list are these:

M058 ALLVASCb mortality ALL VASCULAR DISEASE AGE 0-34 (stand. rate/100,000) (ICD9 390-459, excl 416-7)
M059 ALLVASCc mortality ALL VASCULAR DISEASE AGE 35-69 (stand. rate/1,000) (ICD9 390-459, excl 416-7)
M060 RHEUMHDb mortality RHEUMATIC HEART DISEASE AND FEVER AGE 0-34 (stand. rate/100,000) (ICD9 390-8)
M061 RHEUMHDc mortality RHEUMATIC HEART DISEASE AND FEVER AGE 35-69 (stand. rate/100,000) (ICD9 390-8)
M062 HYPTENSc mortality HYPERTENSIVE DISEASE AGE 35-69 (stand. rate/100,000) (ICD9 401-5)
M063 IHDc mortality ISCHAEMIC HEART DISEASE AGE 35-69 (stand. rate/100,000) (ICD9 410-4)
M064 STROKEb mortality CEREBROVASCULAR DISEASE AGE 0-34 (stand. rate/100,000) (ICD9 430-8)
M065 STROKEc mortality CEREBROVASCULAR DISEASE AGE 35-69 (stand. rate/100,000) (ICD9 430-8)
M066 VASC-STRb mortality ALL VASCULAR DISEASE EXCEPT STROKE AGE 0-34 (stand. rate/100,000) (ICD9 390-459, excl 430-8 & 416-7)
M067 VASC-STRc mortality ALL VASCULAR DISEASE EXCEPT STROKE AGE 35-69 (stand. rate/100,000) (ICD9 390-459, excl 430-8 & 416-7)
M072 COPDc mortality CHRONIC OBSTRUCTIVE PULMONARY DISEASE AND PULMONARY HEART DISEASE AGE 35-69 (stand. rate/100,000) (ICD9 490-6, 416-7)
M112 CONGENHDa mortality CONGENITAL HEART DISEASE AGE 0-4 (cumulative rate/1,000 by age 5) (ICD9 745-6)

Our input items are now these:

M058,M059,M060,M061,M062,M063,M064,M065,M066,M067,M072,M112

Press CRTL+m again and copy/paste the line above into the first input box. Press OK. Again, minimum correlation of 0.5 and associations from different groups. We get 113 associations for heart mortality related items. These can be printed in 3 A4 pages.

File: china-project-heart-corr05-different.pdf (97 kb)
CHINA PROJECT: Heart disease mortality related items: 113 associations, absolute correlation above 0,5.

 

CHINA PROJECT cancer mortality

Regarding cancer mortality, the related items are these:

M022 ALLCAb mortality ALL MALIGNANT NEOPLASMS AGE 0-34 (stand. rate/100,000) (ICD9 140-208)
M023 ALLCAc mortality ALL MALIGNANT NEOPLASMS AGE 35-69 (stand. rate/1,000) (ICD9 140-208)
M024 MOUTHCAc mortality MOUTH CANCER EXCLUDING PHARYNX AGE 35-69 (stand. rate/100,000) (ICD9 140-5)
M025 NASOPCAc mortality NASOPHARYNX AND OTHER PHARYNX CANCER AGE 35-69 (stand. rate/100,000) (ICD9 146-9)
M026 NPConlyc mortality NASOPHARYNGEAL CANCER AGE 35-69 (stand. rate/100,000) (ICD9 147)
M027 OESOPHCAc mortality OESOPHAGEAL CANCER AGE 35-69 (stand. rate/100,000) (ICD9 150)
M028 STOMCAc mortality STOMACH CANCER AGE 35-69 (stand. rate/100,000) (ICD9 151)
M029 COLRECCAc mortality COLORECTAL CANCER AGE 35-69 (stand. rate/100,000) (ICD9 153-4)
M030 LIVERCAb mortality LIVER CANCER AGE 0-34 (stand. rate/100,000) (ICD9 155)
M031 LIVERCAc mortality LIVER CANCER AGE 35-69 (stand. rate/100,000) (ICD9 155)
M032 PANCRSCAc mortality PANCREAS CANCER AGE 35-69 (stand. rate/100,000) (ICD9 157)
M033 BLADDCAc mortality BLADDER CANCER AGE 35-69 (stand. rate/100,000) (ICD9 188)
M034 LARYNXCAc mortality LARYNX CANCER AGE 35-69 (stand. rate/100,000) (ICD9 161)
M035 LUNGCAmc mortality MALE LUNG CANCER AGE 35-69 (stand. rate/100,000) (ICD9 162)
M036 LUNGCAfc mortality FEMALE LUNG CANCER AGE 35-69 (stand. rate/100,000) (ICD9 162)
M037 BREASTCAc mortality FEMALE BREAST CANCER AGE 35-69 (stand. rate/100,000) (ICD9 174)
M038 CERVIXCAc mortality CERVIX CANCER (incl. ca uterus, part unspecified) AGE 35-69 (stand. rate/100,000) (ICD9 179-80)
M039 BRAINCAc mortality BRAIN TUMOUR (MALIGNANT OR NOT) AGE 35-69 (stand. rate/100,000) (ICD9 191, 225.0, 237.5, 239.6)
M040 LYMPHOMAc mortality LYMPHOMA AND MYELOMA AGE 35-69 (stand. rate/100,000) (ICD9 200-3)
M041 LEUKEMIAb mortality LEUKAEMIA AGE 0-34 (stand. rate/100,000) (ICD9 204-8)
M042 LEUKEMIAc mortality LEUKAEMIA AGE 35-69 (stand. rate/100,000) (ICD9 204-8)

The inputs for our macro routine are:

M022,M023,M024,M025,M026,M027,M028,M029,M030,M031,M032,
M033,M034,M035,M036,M037,M038,M039,M040,M041,M042

Press CRT+m and copy/paste the text to the input box (but don’t forget to remove the line break after item M032 before you paste this line). Press OK. Choose minimum correlation 0.5 and associations from different groups. We only have 63 associations for cancer related items. These can be printed in 2 A4 pages. These results will look like this. Don’t forget to check these results (and also the results from our previous example) against the oficial mortality correlations data. For the reasons I explain in the end of this article, there might some differences with the official data, but I don’t think these should invalidate any of the results obtained here.

File: china-project-cancer-corr05-different.pdf (84 kb)
CHINA PROJECT: Cancer mortality related items: 63 associations, absolute correlation above 0,5.

 

CHINA PROJECT blood lipids and saturated fat

Another example, with my favourite items, al things related to cholesterol, saturated fats and omega-3 and omega-6 fatty acids. The items list has all these important variables:

P001 TOTCHOL plasma TOTAL CHOLESTEROL (mg/dL)
D085 CHOL diet survey CHOLESTEROL INTAKE (mg/day/reference man)
P002 HDLCHOL plasma HIGH DENSITY LIPOPROTEIN CHOLESTEROL (mg/dL)
P003 NONHDL plasma NON-HDL CHOLESTEROL (mg/dL)
P004 APOA1 plasma APOLIPOPROTEIN A1 (mg/dL) (non-pooled analysis)
P005 APOB plasma APOLIPOPROTEIN B (mg/dL) (non-pooled analysis)

D084 SATFA diet survey SATURATED FATTY ACID INTAKE (g/day/reference man)
R003 SATFA red blood cell TOTAL LIPID SATURATES (14:0+16:0+18:0+20:0+22:0+24:0) (% of total fatty acid by weight)
R008 P/S red blood cell TOTAL LIPID POLYUNSATURATES/SATURATES (P:S RATIO) (18:2(6)+18:3(3)+20:2(6)+20:3(6)+20:4(6)+20:5(3)+22:6(3))/(14:0+16:0+18:0+20:0+22:0+24:0)

D092 TOTn3 diet survey TOTAL n3 POLYUNSATURATED FATTY ACID INTAKE (g/day/reference man)
D093 TOTn6 diet survey TOTAL n6 POLYUNSATURATED FATTY ACID INTAKE (g/day/reference man)
R005 TOTn6 red blood cell TOTAL LIPID n6 POLYUNSATURATES (18:2+20:2+20:3+20:4) (% of total fatty acid by weight)
R006 TOTn3 red blood cell TOTAL LIPID n3 POLYUNSATURATES (18:3+20:5+22:6) (% of total fatty acid by weight)

Here are our items/variables:

P001,D085,P002,P003,P004,P005,D084,R003,R008,D092,D093,
R005,R006

For minimum correlation of 0.5 and associations of items from different groups, we get 415 results that fit in 11 A4 pages. The corresponding list of associations can be found in this file. Again, you can check these results against the oficial laboratory measurements data. If you find any errrors or inconsistencies please let me know so that I can try to correct them.

File: china-project-blood-corr05-different.pdf (167 kb)
CHINA PROJECT: Cholesterol/blood lipids, saturated fat and omegas-3 & 6: 415 associations, items from different groups, absolute correlation above 0,5.

 

CHINA PROJECT animal and vegetal food

I guess these might be the more interesting associations for those interested in carnivorism vs vegetarianism. So here we go, the CHINA PROJECT items related to animal and vegetal foods/protein/fats are a lot. So I will not spare any paper here, let’s see how all those variables relate to the other items. The animal/vegetal items I would select are all these:

D007 %ANPRKCAL diet survey PERCENTAGE OF CALORIC INTAKE FROM ANIMAL PROTEIN (for reference man)
D008 %PLPRKCAL diet survey PERCENTAGE OF CALORIC INTAKE FROM PLANT PROTEIN (for reference man)
D034 ANIMPROT diet survey ANIMAL PROTEIN INTAKE (g/day/reference man)
D029 ANIMFOOD diet survey ANIMAL FOOD INTAKE (g/day/reference man)
D032 %ANIMFOOD diet survey PERCENTAGE ANIMAL FOOD INTAKE (for reference man)
D036 %ANIMPROT diet survey PERCENTAGE ANIMAL PROTEIN INTAKE (for reference man)
Q168 dANIMFAT questionnaire DAILY CONSUMPTION OF ANIMAL FAT (g/day)

D047 MILK diet survey MILK AND DAIRY PRODUCTS INTAKE (g/day/reference man, as-consumed basis)
D048 EGGS diet survey EGG INTAKE (g/day/reference man, as-consumed basis)
D049 MEAT diet survey MEAT INTAKE (red meat and poultry) (g/day/reference man, as-consumed basis)
D050 REDMEAT diet survey RED MEAT (pork, beef, mutton) INTAKE (g/day/reference man, as-consumed basis)
D051 POULTRY diet survey POULTRY INTAKE (g/day/reference man, as-consumed basis)
D052 FISH diet survey FISH INTAKE (g/day/reference man, as-consumed basis)
D053 ANIMFAT diet survey ADDED ANIMAL FAT (for cooking, spreading etc) INTAKE (g/day/reference man)
Q174 dFISH questionnaire DAYS PER YEAR EAT FISH
Q175 dMEAT questionnaire DAYS PER YEAR EAT MEAT
Q176 dEGGS questionnaire DAYS PER YEAR EAT EGGS
Q177 dMILK questionnaire DAYS PER YEAR CONSUME MILK OR DAIRY PRODUCTS

D028 PLNTFOOD diet survey PLANT FOOD INTAKE (g/day/reference man)
D031 %PLNTFOOD diet survey PERCENTAGE PLANT FOOD INTAKE (for reference man)
D033 PLNTPROT diet survey PLANT PROTEIN INTAKE (g/day/reference man)
D035 %PLNTPROT diet survey PERCENTAGE PLANT PROTEIN INTAKE (for reference man)
Q169 dVEGFAT questionnaire DAILY CONSUMPTION OF VEGETABLE FAT (g/day)
Q172 dGRNVEG questionnaire DAYS PER YEAR EAT GREEN VEGETABLES
Q173 dFRUIT questionnaire DAYS PER YEAR EAT FRUIT

D037 RICE diet survey RICE INTAKE (g/day/reference man, air-dry basis)
D038 WHTFLOUR diet survey WHEAT FLOUR INTAKE (g/day/reference man, air-dry basis)
D039 OTHCEREAL diet survey OTHER CEREAL INTAKE (g/day/reference man, air-dry basis)
D040 STCHTUBER diet survey STARCHY TUBER INTAKE (g/day/reference man, fresh weight)
D041 LEGUME diet survey LEGUME AND LEGUME PRODUCT INTAKE (g/day/reference man, fresh weight)
D042 LIGHTVEG diet survey LIGHT COLOURED VEGETABLE INTAKE (g/day/reference man, fresh weight)
D043 GREENVEG diet survey GREEN VEGETABLE INTAKE (g/day/reference man, fresh weight)
D044 SALTVEG diet survey DRIED AND SALT-PRESERVED VEGETABLE INTAKE (g/day/reference man, as-consumed basis)
D045 FRUIT diet survey FRUIT INTAKE (g/day/reference man, fresh weight)
D054 VEGOIL diet survey ADDED VEGETABLE OIL (for cooking etc) INTAKE (g/day/reference man)

Here are our animal/vegetal items/variables:

D007,D008,D034,D029,D032,D036,Q168,D047,D048,D049,D050,
D051,D052,D053,Q174,Q175,Q176,Q177,D028,D031,D033,D035,
Q169,Q172,Q173,D037,D038,D039,D040,D041,D042,D043,D044,
D045,D054

For minimum correlation 0.5 and associations of items from different groups, we get 628 results that fit in 16 A4 pages. The corresponding list of associations is in this file.

File: china-project-animveg-corr05-different.pdf (165 kb)
CHINA PROJECT: Animal and vegetal food/protein/fats: 628 associations, items from different groups, absolute correlation above 0,5.

Figure: Newspaper reading vs Total Cholesterol.

CHINA PROJECT Newspaper reading and cholesterol levels

I hope this Excel database is useful to those interested in understading more about diet, lifestyle and health. I decided to build it after reading Denise Minger’s articles about the CHINA STUDY. I must say that I find Denise briliant, a briliant mind. Have you already read her last paper The China Study: A Formal Analysis and Response? Since I was also interested in seeing the real numbers, and the original data is fully available on-line, I converted it to a friendly Excel format and did my own calculations. I tried my best to avoid any errors, but be advised that there might be some errors that I’m not aware of. I’ve checked some of the correlations I found (see some comparative examples here), and many of them are absolutely equal to those published in the oficial documents. A few of them, only those related with the mortality items, are not exactly the same value, I suppose because of the different data/year sources used. I used only the year of 1989 (except for a few points from 1975) and, whenever possible, I’ve allways included Taiwan in the calculations. Taiwan adds 16 data points which I think are useful. From what I understand, oficial correlations for laboratory, diet, mortality and questionaire don’t always used Taiwan data (You can read "mainland only" in many of the correlation headers of the oficial correlations). And they also sometimes merge data from different years, from both surveys/years of 1983 and 89, and even from 1975. This might explain the differences I obtained in some of the few mortality item comparisons I tried. Despite of this, I believe this database is overall quite accurate but, anyway, allways check yourself the correlations obtained against the oficial data and see if you can find any relevant inconsistencies. If you find some, please report them and I’ll try to correct them. Since the debate about the CHINA PROJECT data interpretation is just beginning, I hope this data proves useful to everyone, specially the vegetarians at 30 Bananas a Day who are organising their own on-line statistics course. Also, Dr. Colin Campbell took a bit of his time to give us a primer on statistics. I’m reading all these articles and learning a lot from them. Regarding the Excel charts, there are still some problems with them. Automatic charting with Excel is difficult, I had to write some VBA code that is not fully working yet. I hope I can fix this in the next couple of weeks. Any way, I will update this post giving notice when the charting feature is fully functional. The most important part is already fine, the data filtering macro I described above.

These discussions about the CHINA STUDY statistics, specially those on Ned Kocks blog posts & comments, are realy educational for people without formal knowledge on epidemiology or statistics like myself, so don’t miss them if you enjoy learning about these topics. Remember that this is just unadjusted epidemiological data, so don’t jump into conclusions wihout support of intervention data/studies. Just to give you an example, do you know which item from the CHINA PROJECT best relates to total blood cholesterol (P001)? It’s the percentage of people who read the newspaper several times a week (Q027). Actualy, reading the newspaper a lot is a very important (causal?) parameter that has a strong positive, and extremely significant (p

Q027(89) dREADNEWS - questionnaire PERCENTAGE WHO READ THE NEWSPAPER SEVERAL TIMES PER WEEK:

+86 20:5n3 - red blood cell TOTAL LIPID EICOSAPENTAENOIC ACID (20:5(3)) (% of total fatty acid by weight)
+85 COT/cre - urine COTININE (ug/mg creatinine)
+85 AFM1/cre - urine AFLATOXIN M1 (pg/mg creatinine)
+85 TOTCHOL - plasma TOTAL CHOLESTEROL (mg/dL)
+84 NONHDL - plasma NON-HDL CHOLESTEROL (mg/dL)
+84 B2-MGLOB - plasma BETA-2-MICROGLOBULIN (ug/mL)
+84 RIBOFDEF - red blood cell RIBOFLAVIN DEFICIENCY (glutathione reductase activity coefficient)
+84 14:00 - red blood cell TOTAL LIPID MYRISTIC ACID (14:0) (% of total fatty acid by weight)
+83 APOB - plasma APOLIPOPROTEIN B (mg/dL) (non-pooled analysis)
+83 DIABETESc - mortality DIABETES AGE 35-69 (stand. rate/100,000) (ICD9 250)
+83 HBsAg - plasma HEPATITIS B SURFACE ANTIGEN (% of individual samples that were positive; non-pooled analysis)
+82 UREA/cre - urine UREA NITROGEN (mg/mg creatinine)
+81 22:1n9 - red blood cell TOTAL LIPID ERUCIC ACID (22:1(9)) (% of total fatty acid by weight)
+81 NO3mn - nitrosamine study NITRATE (g excreted in 12 hours) (Mean of amounts excreted after ingesting 500mg L-proline with and without 200mg ascorbic acid)
+81 TOTn3 - red blood cell TOTAL LIPID n3 POLYUNSATURATES (18:3+20:5+22:6) (% of total fatty acid by weight)
+80 ENDOCRINc - mortality ENDOCRINE, NUTRITIONAL AND METABOLIC DISEASE AGE 35-69 (stand. rate/100,000) (ICD9 240-279)
+79 Hb - red blood cell HAEMOGLOBIN (g/dL of whole blood) (non-pooled analysis)
+75 20:1n9 - red blood cell TOTAL LIPID GONDOIC ACID (20:1(9)) (% of total fatty acid by weight)
+72 Se - plasma SELENIUM (ug/dL)
+69 18:3n3 - red blood cell TOTAL LIPID LINOLENIC ACID (18:3(3)) (% of total fatty acid by weight)
+68 Fe - plasma IRON (mg/dL)
+68 Zn - plasma ZINC (mg/dL)
+66 ROADACCc - mortality ROAD VEHICLE ACCIDENT AGE 35-69 (stand. rate/100,000) (ICD9 E810-E829)
+65 %FATKCAL - diet survey PERCENTAGE OF CALORIC INTAKE FROM FAT (for reference man)
+65 ANIMFOOD - diet survey ANIMAL FOOD INTAKE (g/day/reference man)
+65 TIBC - plasma TOTAL IRON BINDING CAPACITY (ug/dL)
+65 %ANIMFOOD - diet survey PERCENTAGE ANIMAL FOOD INTAKE (for reference man)

And a strong negative, also very significant, association with these items:

Q027(89) dREADNEWS - questionnaire PERCENTAGE WHO READ THE NEWSPAPER SEVERAL TIMES PER WEEK:

-86 R016(89) - 18:1n9 - red blood cell TOTAL LIPID OLEIC ACID (18:1(9)) (% of total fatty acid by weight)
-85 P047(89) - COTIN>20m - plasma PERCENT OF MALES WITH COTININE >20 ng/mL
-83 R004(89) - MUFA - red blood cell TOTAL LIPID MONOUNSATURATES (16:1+18:1+20:1+22:1+24:1) (% of total fatty acid by weight)
-83 R003(89) - SATFA - red blood cell TOTAL LIPID SATURATES (14:0+16:0+18:0+20:0+22:0+24:0) (% of total fatty acid by weight)
-77 P044(89) - HPYLORI - plasma HELICOBACTER PYLORI IgG ANTIBODY (using cut-off 300) (% of individual samples that were positive; non-pooled analysis)
-74 R007(89) - PUFA - red blood cell TOTAL LIPID POLYUNSATURATES (% of total fatty acid by weight) (18:2(6)+18:3(3)+20:2(6)+20:3(6)+20:4(6)+20:5(3)+22:6(3))
-72 U013(89) - VOLURINEa - nitrosamine study URINE VOLUME (ml excreted in 12 hours after ingesting 500mg L-proline and 200mg ascorbic acid)
-70 U012(89) - VOLURINE - nitrosamine study URINE VOLUME (ml excreted in 12 hours after ingesting 500mg L-proline)
-67 D009(89) - %CARBKCAL - diet survey PERCENTAGE OF CALORIC INTAKE FROM CARBOHYDRATE (for reference man)
-65 D031(89) - %PLNTFOOD - diet survey PERCENTAGE PLANT FOOD INTAKE (for reference man)

Certainly you dont’ beleive there is causality involved in any of these associations. Simply reading the newspaper will not make our cholesterol go up, I suppose. But it may well promote diabetes if you spend the whole day on the sofa reading the newspaper and eating junk food at the same time. So, maybe reading the newspaper is just an excelent marker (epidemiologists call these variables "confounders") of a more urbanized, sedentary and/or industrialised lifestyle. People from Taiwan have the higher cholesterol levels. This is just an example of how any of the raw correlations presented in this post can not be used to directly draw any causal conclusions. Establishing them should be just the first step in an observational study to evaluate which items/variables could be more or less important to explain certain conditions. To find more solid associtations, proper statistical methods shall be applied in order to adjust for the epidemiological confounding factors. And to justify causality between any two items, a biological/biochemical mechanism must be clearly identified and described with support of adequate intervention studies. If you are new to Epidemiology, like I am, try not to forget this ideia: association doesn’t imply causality.

Geographic study of mortality, biochemistry, diet and lifestyle in rural China

"Implications: The chief purpose [of the CHINA PROJECT] is to describe the wide range of differences between different counties in lifestyles and disease-specific mortality rates in rural China, rather than to analyse differences between counties in search of direct evidence of causes. A few of the "ecological" (i.e. geographic) correlations of particular factors with particular diseases do yield good evidence of causality, but the real importance of this study is purely descriptive: better appreciation of the extraordinarily wide range of lifestyles and of disease rates across different Chinese counties will lead to more specific studies."

 

Source: Geographic study of mortality, biochemistry, diet and lifestyle in rural China.

 

CHINA PROJECT/STUDY related links:

Oficial documents:
Study description and methods
Summary (simple) statistics for all 639 variables
Statistics/correlations: Mortality, laboratory, diet and questionaire
Questionaire: Full listings of the six questionnaires
ANNEX: Age-specific deaths and death rates in urban and rural China

Photo: Denise Minger and Dr. Colin Campbell.

Denise Minger:
The China Study, Wheat, and Heart Disease; Oh My! (02-09-2010)
The China Study: A Formal Analysis and Response (pdf) (03-08-2010)
The China Study: My Response to Campbell (16-07-2010)
The China Study: Fact or Fallacy? (07-07-2010)
The China Study (all CHINA STUDY articles)

Dr. Colin Campbell:
China Study Critique
Denise Minger Reply, Campbell Coaliton (or pdf format) (21-07-2010)
A Challenge and Response to The China Study (12-07-2010)
The China Study: Revealing the Relationship between Diet and Disease
A Primer on Statistics

30 Bananas a Day:
Debunking The China Study Critics Discussions (no debunking until now)
My response to Denise’s acceptance of my offer to assist (12-07-2010)
Veganmama’s responses to Denise’s flawed study (16-07-2010)

VegSource:
China Study author Colin Campbell slaps down critic Denise Minger (21-07-2010)

Amazon:
Diet, Lifestyle and Mortality in China: A Study of the Characteristics of 65 Chinese Counties (1990)
The China Study: The Most Comprehensive Study of Nutrition Ever Conducted (book)
What Colin Campbell Doesn’t Tell You About the China Study
Why T. Colin Campbell’s Book is Extremely Misleading

Whole Health Source (Dr. Stephan Guyenet):
The China Study on Wheat (02-09-2010)
Minger Responds to Campbell (17-07-2010)
China Study Problems of Interpretation (08-07-2010)

Weston A. Price Foundation (Chris Masterjohn):
Reductionism and Holism Go Hand in Hand (14/08/2010)
Denise Minger’s Refutation of Campbell’s "China Study" Generates Continued Debate (22-07-2010)
Denise Minger Refutes the China Study Once and For All (13-07-2010)

Health Correlator (Ned Kock):
The China Study II: Cholesterol seems to protect against cardiovascular disease (08-09-2010)
The China Study one more time: Are raw plant foods giving people cancer? (24-07-2010)
The China Study again: A multivariate analysis suggesting that schistosomiasis rules (22-07-2010)
The China Study: With a large enough sample, anything is significant (14-07-2010)

Science Based Medicine:
The China Study Revisited: New Analysis of Raw Data Doesn’t Support Vegetarian Ideology (20-07-2010)

Protein Power (Dr. Michael Eades):
The China Study vs the China study (27-07-2010)

Herectic (Stan Bleszynski):
China Study says wheat is associated with vascular disease (03-08-2010)
China Study says animal fat is healthy! (03-08-2010)
China Study - Raw Data - more plant food = more heart disease! (11-07-2010)

Feasting on Fitness (Kristy A.):
The Study Everyone Talks About Part 1: Correlation is NOT Causation (16-04-2010)
The Study Everyone Talks About: Part 2: The Ravaging Reviews (05-05-2010)

Enzimato:
O Estudo da China (??-07-2010)