search this site.

EPIDEMIOLOGICAL RESEARCH BASED ON LARGE DATA ANALYSIS: STUDY CHARACTERISTICS

Copyright 1995-2023 © by Prof Omar Hasan Kasule - Admin
Category:
Print Friendly and PDFPrint Friendly


Epidemiological Research Based on Large Data Analysis: study characteristics
Omar Hasan Kasule Sr1 and Muhd Ayub Sidiq1
1Institute of Medicine, Universiti Brunei Darussalam

---------------------------------------------------------------------------------------------
Abstract
The paper shows a tendency to publish large size studies based on previously or routinely collected data. Analysis of studies published in one volume of each of 3 major epidemiological studies revealed median study sizes above 1000 for all types of study design, data collection, and study population. Cross sectional and follow up studies had the highest median study sizes when they were based on previously and routinely collected data. The paper discusses some of the problems associated with large studies.
-----------------------------------------------------------------------------------------------------------

Introduction
This study was motivated by the observation that recent epidemiological research is based on existing large data-bases or defined cohorts and that 100% sampling was the usual practice. This is a reversal of the traditional epidemiological practice of selecting a small probability sample from a study population in order to reach conclusion about the target population[1].

Developments in information technology and mass access to the internet opened up new fields of endeavor for the epidemiologist. For example data could be collected from a large number of people using internet-based questionnaires [2]

The objective of the research was to survey epidemiological research published in 2006 in three high-impact journals to make a statistical description of the characteristics of these studies. The three journals selected for study were the American Journal of Epidemiology 2006 Volume 169 Nos. 1-12, The International Journal of Epidemiology 2006 Volume 35 Nos. 1-4, and The European Journal of Epidemiology 2006 Volume 21 Nos. 1-9.

Methods
The study included only original papers that involved raw data. Reviews, meta-analyses, and analyses based on published data were excluded. A pre-tested data abstract form was used to abstract the following essential information from each original research article: title, authors, issue and volume number, date of publication, type of study (cross sectional, case control, cohort, randomized community control, randomized clinical), Study population (general population, defined population, ongoing study), type of data collection (routinely collected data, newly collected data, previously collected data or a combination among the above) and total number of study subjects (number recruited before any exclusions). Defined populations were hospitals, health insurance of health maintenance organizations, clinics, schools, factories, and ongoing studies. For case control studies cases and controls were added up. For ongoing studies it was assumed that data collection was new unless a special mention was made of using previously collected data. The data was keyed into an SPSS data base for categorical analysis using the chi-square test statistic to test for association. The Kruskall-Wallis non-parametric test statistic was used to compare study size by various study designs and types of data collection.

Results
A total of 137 studies were analyzed. Table 1 shows a significant variation in journal preference for study designs and methods of data collection. There was however no significant variation among journals in the choice between defined and general study populations. Table 2 shows that the median study side was over 1000 for all journals and types of study design. Its variation among journals was significant for follow up studies and not cross sectional and case control studies. The median study size did not vary significantly among the 3 journals for different study populations and methods of data collection. Table 3 shows that cross sectional and follow up studies had significantly higher median study size in general populations than in defined populations. No such significant variation was seen on case control studies. Table 4 shows that follow up studies had higher median study size if based on previously and routinely collected data. No such significant variation was observed for cross sectional and case control study designs. Table 5 shows significant variation of median study size with the type of data collection. Study size in defined populations was highest for newly and routinely collected data whereas in the general population median study size was highest for previously and routinely collected data.

Discussion
Median study sizes were highest for cross sectional and follow up studies and when based on previously or routinely collected data. This was due to availability of large data bases with routinely or previously collected information. The availability of large data bases and high speed computers encouraged epidemiologists to analyze data without probability sampling. Several issues arise in discussing large size studies [1]. A large data could give very stable parameters but the same degree of precision could have been obtained from a smaller sample. What was the lost was the ability of the epidemiologist to inspect a small manageable data set, internalize it, and let his intuition act before the data is analyzed. The more intimate contact of the epidemiologist with the data traditionally accounted for deep understanding and discussion which are missed in the new trend. Easy availability of large databases also encouraged epidemiologists to plunge into data analysis before serious thought about the research questions. In some cases the research questions could be prompted by preliminary analysis which can lead to numerous biases. Use of large data sets had the advantage of external validity which had never been the primary objective of epidemiological research. Epidemiologists traditionally aimed at carrying out a small study based on probability sampling so that they could easily identify and control confounding and other sources of bias with the ultimate aim of internal validity. They knew that external validity (generalization) would be attained inductively by consideration of several studies that are internally valid. Use of large sets of routinely collected data also raised the issue of the quality of the data which is collected with service and administrative and not research considerations in mind.
           

TABLE 1: CLASSFICATION OF ARTICLES BY JOURNAL
Study Characteristics
IJE
EJE
AJE
Total
X2(df)
P valuea
n
(%)
n
(%)
n
(%)
Study design
Cross sectional
16
(39.0)
33
(60.0)
13
(31.7)
62
10.10 (4)
0.039
Case control
9
(22.0)
9
(16.4)
7
(17.1)
25
Follow up
16
(39.0)
13
(23.6)
21
(51.2)
50











Study population
Defined population
12
(29.3)
23
(41.8)
22
(53.7)
57
5.02 (2)
0.081
General population
29
(70.7)
32
(58.2)
19
(46.3)
80











Data Collection
Newly & routinely
6
(14.6)
5
(  9.1)
3
(  7.3)
14

0.030b
Newly
13
(31.7)
28
(50.9)
25
(61.0)
66

Previously
0
(  0.0)
1
(  1.8)
3
(  7.3)
4

Routinely
22
(53.7)
21
(38.2)
10
(24.4)
53

a Chi-square test;  b Fisher’s exact test;
IJE = International Journal of Epidemiology
EJE = European Journal of Epidemiology
AJE = American Journal of Epidemiology


TABLE 2: NUMBER OF RESEARCH SUBJECTS BY STUDY CHARACTERISTICS AND JOURNAL
Study characteristics
Journal
n
Number of research subjects
X2.(df)
P valuea
Median
Min.
Max.
Study design







Cross sectional
IJE
16
7,183
107
212,467,094
0.63 (2)
0.730
EJE
33
4,599
112
36,000,000
AJE
13
2,255
139
212,467,094








Case control
IJE
9
1,263
288
166,310
1.61 (2)
0.446
EJE
9
1,051
372
2,222,404
AJE
7
1,875
730
212,467,094








Follow up
IJE
16
60,925
1,016
60,000,000
7.80 (2)
0.020
EJE
13
9,778
34
11,000,000
AJE
21
2,446
209
212,467,094








Study population







Defined population
IJE
12
6,381
726
246,146
2.47 (2)
0.291
EJE
23
1,272
34
6,240,130
AJE
22
2,102
299
1,299,177








General population
IJE
29
14,495
107
212,467,094
1.36 (2)
0.506
EJE
32
7,404
188
36,000,000
AJE
19
10,932
139
212,467,094








Data Collection







Newly & routinely
IJE
6
2,380
274
246,146
0.75 (2)
0.686
EJE
5
11,081
1,051
2,222,404
AJE
3
11,234
1,068
212,467,094








Newly
IJE
13
3,290
288
14,495
3.90 (2)
0.142
EJE
28
937
34
6,240,130
AJE
25
2,010
139
21,610








Previously
IJE
-
-
-
-
0.20 (1)
0.655
EJE
1
56,214
56,214
56,214
AJE
3
1,516
619
212,467,094








Routinely
IJE
22
180,155
107
212,467,094
2.43 (2)
0.296
EJE
21
19,801
212
36,000,000
AJE
10
399,910
1,832
212,467,094
a Kruskal-Wallis Test
IJE = International Journal of Epidemiology
EJE = European Journal of Epidemiology
AJE = American Journal of Epidemiology


TABLE 3: NUMBER OF RESEARCH SUBJECTS BY STUDY DESIGN AND STUDY POPULATION
Study design
Study pop.
n
Number of research subjects
Z
P valuea
Median
Min
Max








Cross sectional
Defined pop.
22
2,222
112
6,240,130
-2.28
0.023
General pop.
40
10,832
107
212,467,094








Case control
Defined pop.
10
1,552
726
2,222,404
-1.00
0.318
General pop.
15
1,083
288
212,467,094








Follow up
Defined pop.
25
2,311
34
1,299,177
-3.77
<0.001
General pop.
25
83,875
188
212,467,094
a Mann-Whitney test


TABLE 4: NUMBER OF RESEARCH SUBJECTS BY STUDY DESIGN AND TYPE OF DATA COLLECTION
Study design
Data collection
n
Number of research subjects
X2.(df)
P valuea
Median
Min
Max








Cross sectional
New & routine
3
7,000
274
11,193
7.49 (3)
0.058
Newly
33
2,650
112
6,240,130
Previously
3
1,516
619
212,467,094
Routinely
23
95,000
107
212,467,094








Case control
New & routine
7
1,678
1,051
212,467,094
2.86 (2)
0.240
Newly
13
909
288
4,778
Previously
-
-
-
-
Routinely
5
1,272
828
166,310








Follow up
New & routine
40
15,127
11,081
246,146
29.53 (3)
<0.001
Newly
20
995
34
11,267
Previously
1
56,214
56,214
56,214
Routinely
25
87,922
212
212,467,094
a Kruskal-Wallis Test


TABLE 5: NUMBER OF RESEARCH SUBJECTS BY STUDY POPULATION AND METHOD OF DATA COLLECTION
Study population
Data collection
n
Number of research subjects
X2 (df)
P valuea
Median
Min
Max








Defined population
New & routine
7
11,234
7,000
2,222,404
12.10 (2)
0.002
Newly
39
2,010
34
6,240,130
Previously
-
-
-
-
Routinely
11
1,272
212
1,299,177








General population
New & routine
7
1,263
274
212,467,094
27.47 (3)
<0.001
Newly
27
1,653
139
47,859
Previously
4
28,865
619
212,467,094
Routinely
42
187,530
107
212,467,094
a Kruskal-Wallis Test


REFERENCES
1. Omar Hasan Kasule. The transition from sample to population epidemiology. Journal of the University of Malaya Medical Center (in press)
2. Alexandra Ekman, Paul W Dickman, Asa Klint, Elisabete Wederpass, and Jan-Eric Litton. Feasibility of using web-based questionnaires in large population-based epidemiological studies, European Journal of Epidemiology 2006; 21: (2): 103-111.