Microsoft Word search phase 3 Title Page Amendment

6.
 
Statistical Considerations 
6.1.
 
TEMPORAL TRENDS IN INCIDENCE OF DIABETES (Research Questions 1.1 & 
1.2) 
SEARCH has ascertained each new case of T1D and T2D in its surveillance areas since 
2002, allowing us to estimate the incidence of diabetes over time.  We will continue this 
surveillance in SEARCH 3, and a primary objective is to assess the significance of changes 
in the incidence of T1D and T2D over time, overall and by levels of important characteristics 
such as age, sex, and race/ethnicity, and to determine if the changes over time vary by these 
characteristics.   
Chi-square tests will be used to determine if the incidence of diabetes changes over time, 
overall and by characteristics of interest.  The chi-square test is robust against a large range 
of possible differences; however, it is not particularly powerful for detecting specific 
patterns.  While numerous patterns of change are possible (consistently increasing, 
consistently decreasing, increasing then decreasing, decreasing then increasing, etc.), we are 
primarily interested in detecting consistently increasing or decreasing changes over time.
  
The Cochran-Armitage trend test will be used to determine if the changes over time are 
increasing or decreasing.  In a more general framework, weighted linear regression and 
logistic regression will be used to assess changes in diabetes incidence over time.  These 
models allow us to assess the effects of characteristics of interest and to determine if the 
trends differ by levels of the characteristics (e.g., younger vs. older age-groups, NHW vs. 
minority, males vs. females).
 
Approximately 5,000,000 youth less than 20 years of age live in the SEARCH surveillance 
area.  About half are female, 60% NHW, 14% AA, 16% Hispanic, and the rest A/PI or AI. 
Using our initial incidence estimates 
(1)
 and assuming the population remains constant over 
time, we will be able to detect a slope of 0.13 (or 0.7% annual change) in the overall 
incidence of T1D.  This slope is well within the range previously reported in the literature: an 
annual increase of 3.9% (95% CI 3.6-4.2) in Europe 
(2)
 and 2.3% (95% CI 1.6-3.1) in 
Colorado 
(2)
.  Importantly, given the large population and the relatively long period of 
surveillance (13 years), we will have adequate power to detect changes in incidence of T1D 
by age groups even smaller than those previously reported, for example, among Colorado 
youth: 3.5% (2.1-4.9) for age-groups 0-4 years; 2.2% (1.0-3.5) for 5-9 years 1.8% (1.0-2.7) 
for 10-14 years and 2.1% (0.5-3.7) for 15-17 years 
(2)
.  Moreover, we will be able to detect 
reasonable slopes in the incidence of T1D among the major racial/ethnic groups: 0.81% 
annual change for NHW [2.7% (1.9-3.6) previously reported (2); 2.2% for AA; 2.1% for 
Hispanics (1.6% (0.2-3.1) previously reported 
(2)
.  We will also be able to detect a slope of 
0.13 (or 1.35% annual change) in the overall incidence of T2D in youth 10 and older, and 
reasonable slopes by sex, age-group and major racial/ethnic groups.  There are no published 

---

Section 6A - Statistical Considerations (Phase 3 - 11/2010) 
Section 6A - Page 2 
 Registry 
Study
 
 
trends in incidence of T2D in youth, in the U.S. or worldwide, to provide comparison for our 
estimates. 
Regression models will also be used to assess differences in the slopes by demographic 
characteristics (i.e, covariate by time interactions).  Table 6.1 shows the differences in slopes 
of T1D and T2D, by demographic characteristics, which we can detect in 2014 (e.g., males 
vs females, NHW vs minority, and 0-9 vs 10-19 for T1D or 10-14 vs 15- 19 for T2D) with 
80% power at the 5% two-sided level of significance, assuming a variety of slopes.  For 
slopes ranging from 0.2 to 0.5 the detectable differences between groups are almost identical.  
We see that for T1D, using data through 2014, we can detect a difference in slopes of 0.28.  
The detectable differences in the slopes of T2D diabetes are slightly larger because of the 
smaller denominator (youth 10 and older), but not much larger since there is a lower 
incidence of T2D.  Note that the most conservative estimates are shown in Table 6.1 (e.g., 
NHW vs minority for T1D as the denominators differed most for these groups); detectable 
differences were similar for other group comparisons. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6.2.  TEMPORAL CHANGES IN CLINICAL CHARACTERISTICS (Research Question 
1.3) 
Another research question focuses on assessing potential temporal changes in the 
demographic and clinical characteristics of youth with diabetes over time.  Analysis of 
variance (for continuous measures like age and HbA
1c
) and logistic regression (for 
dichotomous measures like prevalence of autoimmunity) will be used to assess changes in 
these outcomes over time.  Time will initially be entered as a categorical variable to check 
for any differences over time.  Time will then be considered continuously to check for 
consistent changes over time (slopes).  Time by covariate interactions will be assessed to 
Table 6.1 - Differences in slopes detectable with 80% 
power in 2014 (between genders, races, and age groups) 
Slope 1* 
Slope 2* 
Difference 
T1D  
 
0.20 0.47 
0.27 
0.30 0.58 
0.28 
0.40 0.68 
0.28 
0.50 0.78 
0.28 
T2D  
 
0.20 0.52 
0.32 
0.30 0.63 
0.33 
0.40 0.74 
0.34 
0.50 0.84 
0.34 
* Slope 1 is the observed slope in one group (e.g., males; 
NHW; etc) and Slope 2 is the observed slope in the other 
group (e.g., females; Other; etc) 

---

Section 6A - Statistical Considerations (Phase 3 - 11/2010) 
Section 6A - Page 3 
 Registry 
Study
 
 
determine if the slopes differ by level of important covariates (e.g., does the change in age 
over time differ for males and females).  In Table 6.2, we show the slopes that can be 
detected with 80% power using the data through 2014, assuming that 1) onset age will be 
available each year and data for the other measures will be available for cohorts incident in 
2002-2006, 2008, and 2012; 2) the sample sizes for onset age and for variables of interest 
among T2D youth will remain constant; and 3) the sample size for variables of interest 
among T1D youth will remain constant between 2002 and 2008 and will be 5/6(N) for 2012 
(due to sampling only 50% of the NHW age <10 years).  For example, there are about 970 
incident T1D cases per year with an IPV in our surveillance area.  The mean age of newly 
diagnosed T1D cases in 2002 was 9.4 years with a standard deviation of 4.7 years.  Using 
data through 2014, we can detect a yearly change in onset age of 0.031 years (e.g., 9.4 in 
2002 vs 9.0 in 2014).  Approximately 84% of T1D had evidence of autoimmunity in 2002.  
With a sample size of approximately 440 for diabetes autoimmunity each year it is collected, 
we’ll be able to detect a change of 0.6% per year (e.g., 84% in 2002 vs 90% in 2012). 
  
Table 6.2 - Changes in Clinical Presentation, By Type, 
Detectable with 80% Power in 2014  
Variable Mean 
(SD)
N per 
year 
Detectable 
change 
T1D  
 
 
   Onset Age (years) 
9.4 (4.7) 
970 
0.031 
   BMI-z 
0.6 (1.0) 
500 
0.015 
   HbA
1c
 (%) 
7.9 (1.6) 
470 
0.025 
   Insulin Sensitivity 
10.3 (3.3) 
430 
0.054 
   FCP (ng/ml) 
0.5 (0.5) 
470 
0.008 
   Autoimmunity (%) 
84% (36%)
440 
0.006 
   DKA (%) 
30% (46%)
760 
0.006 
T2D  
 
 
   Onset Age 
14.3 (2.8) 
250 
0.037 
   BMI_z 
2.1 (0.7) 
85 
0.026 
   HbA
1c
 7.5 
(2.4) 
85 
0.088 
   Insulin Sensitivity 
4.6 (2.8) 
75 
0.109 
   FCP (ng/ml) 
3.3 (1.9) 
85 
0.069 
6.3.
 
INFORMING SUSTAINABLE SURVEILLANCE 
To address this aim, we are proposing to provide consultation and support to the CDC and 
the NIH for the development of low-cost, sustainable public health surveillance systems for 
diabetes in children.  Based on our experience we know that efficient and complete 
ascertainment of cases of T2D in youth, and ascertainment of cases of diabetes in older youth 
present significant challenges.  We are proposing two approaches to better understand the 

---

Section 6A - Statistical Considerations (Phase 3 - 11/2010) 
Section 6A - Page 4 
 Registry 
Study
 
 
underlying systems of care in order to define best practices for case ascertainment.  The first 
builds on data that has been collected over the past 10 years of the SEARCH study as well as 
data to be collected in SEARCH 3; the second compares data collected by one of the existing 
six SEARCH centers - a HMO (KPSC) - with analogous data collected by a large integrated 
multi-payer system (UNC). 
6.4.
 
EVALUATING CASE ASCERTAINMENT STRATEGIES: ANALYSIS OF SEARCH 
REGISTRY DATA 
For research question 2.1, we will evaluate (a) whether a longer period of calendar time is 
required to ascertain cases of youth with T2D or older youth with any type of diabetes than 
youth with T1D and younger youth; (b) whether any calendar time difference in 
ascertainment by diabetes type and age can be accounted for by the time from clinical 
diagnosis to hospitalization or receipt of specialty care services; and (c) if there are temporal 
changes in observed differences in time to case ascertainment.  Date of diagnosis, date of 
birth, sex, clinical diabetes type, type of health care provider (pediatric endocrinologists, 
pediatricians, family practice physicians, adult endocrinologists) and the system in which 
each youth with diabetes received their care (community clinic, university based health 
system, managed health care organization) around the time of diagnosis are obtained from 
the medical record, the IPS, or administrative data sources at the time of case validation.  
Calendar time from date of diagnosis to date of case registration will be calculated in months 
so that time from diagnosis to registration can be compared by type within each category as 
well as within type by age category.   
 
Using this information from SEARCH registered cases we will be able to quantify the timing 
and completeness of case ascertainment under a variety of scenarios.  For example, we can 
simulate specific ascertainment scenarios to compare the completeness of ascertainment of 
youth with T1D versus T2D by age category if a system relied solely on pediatric 
endocrinology and hospital-based surveillance networks as compared to if a system that 
relied on these sources plus primary care providers.  We will identify efficient ways to 
optimize ascertainment of youth with T1D or T2D, using data collected to date by the 
SEARCH study and through the 2011 incident cohort.  This approach goes beyond 
examining variation by study centers and takes advantage of the diversity within and between 
the existing SEARCH study centers which conduct ascertainment state-wide (SC, CO), in 
defined regions (WA, OH), from enrollees in managed health care organizations (CA), and 
from enrollees in IHS facilities. 
 

---

Section 6A - Statistical Considerations (Phase 3 - 11/2010) 
Section 6A - Page 5 
 Registry 
Study
 
 
6.5.
 
MORTALITY ANALYSIS COMPONENT 
Crude mortality rates per 1,000 person-years (p-y) will be calculated, and Poisson and Cox 
proportional hazards regression analyses will be used to evaluate predictors of mortality (e.g., 
DM type, race/ethnicity, insurance type).  Using the mortality rate of 2.50/1,000 p-y from the 
Chicago DM registry cohort study
 (3)
, we estimate that we will have 117 deaths in the 
SEARCH cohort.  We will calculate the Standardized Mortality Ratio (SMR) comparing the 
SEARCH cohort to mortality data from the geographic areas from which the SEARCH 
sample participants are drawn, accounting for the demographic distribution of the cohort with 
DM.  

---

Section 6A - Statistical Considerations (Phase 3 - 11/2010) 
Section 6A - Page 6 
 Registry 
Study
 
 
Reference List 
 
1.
 
Writing Group for the SEARCH for Diabetes in Youth Study Group; Dabelea D, Bell RA, 
D'Agostino RB, Jr., Imperatore G, Johansen JM, Linder B et al. Incidence of diabetes in 
youth in the United States. J.A.M.A. 2007;297:2716-24. 
2.
 
Vehik K, Hamman RF, Lezotte D, Norris JM, Klingensmith G, Bloch C, Rewers M, Dabelea 
D. Increasing incidence of type 1 diabetes in 0- to 17-year-old Colorado youth. Diab.Care 
2007;30:503-9. 
3.
 
Burnet DL, Cooper AJ, Drum ML, Lipton RB. Risk factors for mortality in a diverse cohort 
of patients with childhood-onset diabetes in Chicago. Diabetes Care 2007 Oct;30(10):2559-
63. 
 

---

SEARCH Phase 3 Protocol - Section 6B 
Cohort Study 
Statistical Considerations 
Table of Contents