Evaluation of Small Airways Disease in asthma medical research: Application of Structural Equation Modeling (SEM) and Latent Transition Analysis (LTA)

Thomas ZwingersMarco Pannacci

Agenda

• Study Design

• Small Airways Disease (SAD) in asthma• Structural Equation Model• Cross-sectional Model• Longitudinal Model• Summary

7

Study Design and Objectives

• Study Design: Multinational, multicenter, non-pharmacological intervention study with a cross-sectional and a longitudinal phase.

• Follow-up time : 1 year

• Major objectives:

To evaluate which (combination of) clinical methods best assesses the abnormalities of small airways disease in asthma and their role in the clinical manifestations of asthma, both cross-sectionally and longitudinally.

7

Study Population

No. of AsthmaticPatients

No. of Healthy Volunteers

Screened Set 883 104

Evaluable Set 772 98

• Planned Sample Size: 900 patients (800 asthmatic patients + 100 healthy volunteers).

ASTHMA

7Source: https://en.wikipedia.org/wiki/Asthma

Asthma is a common long-term inflammatory disease of the airways of the

lungs. It is characterized by variable and recurring symptoms, reversible

airflow obstruction, and bronchospasm.

There is no cure for asthma.

Poorly controlled inflammation

in small airways may

contribute to

• asthma exacerbations,

• air trapping,

• lung function decline, and

• irreversible structural

changes

ASTHMA

7

Source: http://ginasthma.org/2018-pocket-guide-for-asthma-management-and-prevention/

Asthma Prevalence

• An estimated 300 million people worldwide suffer from asthma, with 250,000 annual deaths attributed to the disease.

• It is estimated that the number of people with asthma will grow by more than 100 million by 2025.

7

Source: http://ginasthma.org/2018-pocket-guide-for-asthma-management-and-prevention/

Categories of asthma severity

Mild asthma:

Well controlled with SABA or low dose ICS

GINA Steps 1 or 2

Moderate asthma:

Well controlled with low dose ICS/LABA

GINA Step 3

Severe asthma:

requires moderate or high dose ICS/LABA ± add on,

or remains uncontrolled despite this treat

GINA Steps 4/5

ASTHMA

Structural Equation Modeling (SEM)

SEM is a very general statistical modeling method.

It can be viewed as a combination of exploratory factor analysis, confirmatory factor analysis, path analysis or regression and others.

Typical example: measurement of intelligence.

8

• We can’t measure the concept of interest (intelligence, socio-economic status ..) directly

latent variables.

• Hovewer we can refer to observable indicators that are influencedby the latent variable, e.g scores obtained by a series of testsabout linguistic or arithmetic abilities

observed variables.

Structural Equation Modeling (SEM)

Structural Equation Models consist of

a structural model representing the relationship between the latent variables of interest, and

A B

CStructural model

A

I7

I6

Measurement model A

B

I5 I4 I3

Measurement model B

C

I1

I2

Measurement model C

measurement models representing the relationship between the latent variables and their manifest or observable indicators

Specification of SEM model – Cross-sectional phase

The Path Diagram describes the theoretical framework that forms

the basis for specification of our initial model.

The rationale about hypotheses and directionality specifications in

the diagram was built by the statisticians consulting the experts of

the clinical team during two TC meetings.

Observed variables in the Path Diagram are represented with

squares or rectangles; latent variables with circles; hypothesized

directional effects (causal relationships) of one variable on

another, with a line with a single arrowhead; correlations between

variables with a curved line with two arrowheads.

7

Lung physiology

Systemic Inflammation

?

Small/large airways

Inflammation

Subjective status

Asthma history

Small/large airways disease

Age

Sputum induction

FeNOHaematology

Phadiatop test

ACT Total Score

BHQ Total Score

Age at asthma diagnosis

Number of Exacerbations

(last year)

Healthcare resourceconsumption

Smoke history

Duration of disease

Multiple-Breath Nitrogen Washout Test (MBNWT)

Impulse OscillometrySystem (IOS) MarkersSpirometry Markers

Methacholine challenge test

Body plethysmography Markers

ACQ-6 Total Score

EuroQol VAS

Mni Total Score

?? ?

?

Specification of SEM model

Outcome of SEM model: Asthma clusters

Overall the clusters represent GINA severity stages predominantly, suggesting that SAD increases with asthma severity.

0

10

20

30

40

50

60

70

80

90

Asthma Group 1 Asthma Group 2 Asthma Group 3 Asthma Group 4

% o

f p

atie

nts

GINA Steps vs. Asthma Groups

Step 1 Step 2 Step 3 Step 4 + 5

This analysis identifies specific asthma, not SAD, clusters.

Small Large Both Systemic effects Patient-reported outcome% fall in FVC at PC20 R20 (kPa L-1 s-1) PC20 (mg/mL) WBC Asthma Control Test (ACT)

R5-20 (kPa L-1 s-1) FENO single flow (50ml/s)

PD20 (mg) Neutrophils Mini asthma quality of life questionnaire

X5 (kPa L-1 s-1) Sputum PD20 & PC20 categories

Monocytes Asthma control questionnaire (ACQ-6)

AX (Hz kPa L-1 s-1) FEV1 %pred Lymphocytes Bronchial HyperresponsivenessQuestionnaire (BHQ)

Scond*VT in liter FEV1/FVC (L/sec) Eosinophils EuroQol-5D-5LSacin*VT in liter IVC %pred Basophils EuroQol VAS (respondent’s self-rated

health)RV %pred FENO multiple flows

(100, 150, 350, other)Phadiatop test Total number of hospital admission

days in the past 12 months before V1

RV / TLC %pred Number of asthma-specific emergency room or urgent care visits

FRC %pred Number of unscheduled consultations visits due to symptoms worsening

FEF50%/FVC Total number of Unscheduled tests for asthma

FEF25-75%/FVC Number of exacerbation in the last year SAD

only the variables reflecting small airways disease and

both small and large airways disease were considered for

the SAD measurement model

Statistical Analysis Plan - New approach

Study Population

No. of AsthmaticPatients

No. of Healthyvolunteers

Evaluable Set 772 98

SAD SEM Analysis* 761 Not used

*We lost only 6 patients with no data for small airways disease variables and

we excluded 5 patients that were outliers (based on the model indicators for

outliers detection).

Clinical SAD SEM – Cross Sectional Model

Model Fit:Root Mean Squared Error of Approximation(RMSEA): 0.048Comparative Fit Index (CFI): 0.990

RV/TLC % pred

SAD

LUNG1

LUNG2

LUNG3

.619 (.058)

.731 (.047)

.038 (.007)

R5-20

AX

X5

Sacin

FEF50/FVC

FEF25-75/FVC

1.000 (.000)

Scond

.108 (.009)

.183 (.016)

.009 (.001)

.782 (.042)

.070 (.041)

.145 (.039)

.738 (.044)

.981 (.004)

.617 (.047)

.518 (.046)

0.944 (.005)

.904 (.009)

.996 (.000)

.925 (.021)

.512 (.043)

.645 (.058)

.285 (.053)

.964 (.021)

.291 (.056)

.585 (.074)

SAD SEM model based clustering

-1 0 1 2

Clinical SAD SEM Score

0

20

40

60

80

100

Count

21Clinical SAD SEM Groups

Relationship between Clinical SAD (SEM) groups

and GINA classes

Clinical SAD

SCORE

Gina 1 Gina 2 Gina 3 Gina 4 Gina 5

Mean (n)-0.143

(134)

-0.035

(83)

-0.048

(205)

0.071

(296)

0.239

(46)

Anova p-value <.0001

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

Step 1 Step 2 Step 3 Step 4 Step 5

Mean Clinical SAD SCORE

A B

C

A

I7

I6

B

I5 I4 I3

C

I1

I2

Clinical SAD SEM – Longitudinal Model

Latent Transition Analysis

VISIT 1 VISIT 3

A B

C

A

I7

I6

B

I5 I4 I3

C

I1

I2

Clinical SAD SEM – Latent Transition Analysis

VISIT 1 VISIT 2 VISIT 3

SAD Groups G1 = 59%G2 = 41%

SAD Groups G1 = 61%G2 = 39%

SAD Groups G1 = 60%G2 = 40%

14 patients move from G1 to G228 patients move from G2 to G1

14 patients move from G1 to G29 patients move from G2 to G1

6 months 6 months

27 patients move from G1 to G236 patients move from G2 to G1

VISIT 1 VISIT 3

Summary

We were able to define a score that reflects the amount of

physiological small airways impairment

The score associated significantly with measures of asthma

control, disease instability and asthma severity as defined by

GINA stages

SAD appears to be clinically meaningful as suggested by the

association of physiologic parameters of SAD with asthma

severity, asthma control and health care utilization

We can detect asthma subtypes based on small airways

dysfunction

Literature

1. Bollen, K. A. (1989). Structural equations with latent variables. New York, NY: John Wiley & Sons, Inc.

2. Bentler P M 1990 Comparative fit indices in structural models. Psychological Bulletin 107: 238–46

3. Chung, H., Park, Y., & Lanza, S.T. (2005). Latent transition analysis with covariates: pubertal timing and substance use behaviors in adolescent females. Statistics in Medicine, 24, 2895 - 2910.

4. Muthen B 1984 A general structural equation model with dichotomous, ordered categorical, and continuous latent variable indicators. Psychometrika 49: 115–32

5. Stephanie T. Lanza, Linda M. CollinsA : New SAS Procedure for Latent Transition Analysis: Transitions in Dating and Sexual Risk Behavior. Dev Psychol. 2008 March ; 44(2): 446–456

6. William McNulty* and Omar S. Usmani: Techniques of assessing small airways dysfunction. European Clinical Respiratory Journal 2014, 1: 25898