Testing and Treatment for Hepatitis C: When and Where Dr. Julie Swann Georgia Institute of Technology H. Milton Stewart School of Industrial and Systems Engineering Feb 2007

Joint work with Daniel Faissol (GT), Paul Griffin (GT), Susan Griffin (CDC), Eser Kirkizlar (GT) 1

Hepatitis C Background „

Liver disease caused by blood-borne Hepatitis C virus (HCV) ‰

„

Most infected people are asymptomatic for decades ‰

„ „

Liver disease is 10th leading cause of death among US adults 3.9 million people in US are currently infected but many are unaware

Treatments are somewhat effective (~ 50%) Behavior is important to progression and secondary infections

2

1

Disease Taxonomy Human Papillomavirus

Lung Cancer

3

1 and 2

1

2

3

2

2

2

2

3

1

2

2

1

1 or 2**

1

2 2

Chlamydia

Behavior

2

1

Infectious

3

Length

1

Flu

Morbidity

2

1

1 and 3

2 and 3

1, 2, or 3

2 and 3

Population

1 or 2*

2

2

2

2

2

Symptoms

1 or 2

1

1

2

3

2

Treatment

2

1

2

1

1

3

„ „ „ „ „ „ „

Malaria

Hepatitis C

Asthma

3 1 1

Behavior {Impacts transmission (1), Impacts morbidity (2), Has no effect (3)} Infectious {Communicable (1), Transmissible (2), Not Infectious (3)} Length {Chronic (1), Acute (2)} Morbidity {Small effect (1), Large effect (2), Leads to death (3)} Population {Targets specific groups (1), Differs by risk (2), No discrimination (3)} Symptoms {Brief time to symptoms (1), Long time to symptoms (2), Asymptomatic(3)} Treatment {Cure (1), Improves quality of life (2), None (3)} *Genetic linkages are not well understood **HPV can lead to a chronic state (e.g., cervical cancer)

Screening for Disease Presence „ „

Identify populations (at-risk) to test for disease Evaluate cost-effectiveness ‰ ‰ ‰ ‰

„

Prevalence of disease Accuracy and cost of test Progression and cost of disease Cost reduction through disease awareness (e.g., treatment or behavioral change)

(Sometimes) determine frequency or timing 4

2

Hepatitis C Findings „

„

„

„

Treatment of HCV is cost-effective ‰ Kim et al (1997), Salomon et al (2003), Neumann et al (2006), Saab et al (2002) Testing for HCV ‰ Yes for drug users (Castelnuovo et al 2006) or high risk groups (Gordon 1999) ‰ Not for pregnant women (Plunkett and Grobman 2005) or average adults (Singer and Younossi 2001) or post-transfusion patients (Pereira and Sanz 2000) Official recommendations ‰ USPHS & IDSA: HIV positive persons ‰ USPSTF: not general population (no other recommendations) ‰ CDC: High risk (drug users, medical staff) ‰ None comment on timing Our contribution: timing (and frequency), impact of behaviors 5

Other Findings „

„

„

„

Monte Carlo simulation of Markov models for disease progression or screening ‰ Goldie and Kuntz (2003), Rosenquist and Lindfors (1994), Sonnenberg et al (2000) Repeated screenings (usually equally spaced) ‰ Brenner et al (2006), Chen et al (2001), Siebert et al (2003), Paltiel et al (2005) ‰ Kaplan and Satten (2000): prevalence Æ interval ‰ Diehl et al (2006): computational testing Analytical approaches to timing ‰ Machine inspection and replacement (Eckles 1968, Smallwood and Sondik 1973, Ozekici and Pliska 1991, Grosfeld-Nir 1996) ‰ Medical screening (Parmigiani 1993, Zelen 1993) ‰ Timing of liver transplants (Alagoz et al 2004, 2006) Our contribution: analytical analysis of timing of unknown disease with behavioral impact 6

3

Motivation „

Analyze general models for timing of testing decisions ‰ ‰

„

Determine appropriate testing and treatment for Hepatitis C with societal perspective ‰ ‰

„

Disease may be asymptomatic Knowledge can affect behavior (progression or infection)

Minimize cost to the system including productivity losses or Maximize Quality-Adjusted Life Years (QALYs) gained

Use mathematical modeling to inform policymaking

7

Hepatitis C Progression Rx Uninfected

Infected w/out Cirrhosis

Compensated Cirrhosis

Hepatocellular carcinoma

Decompensated Cirrhosis $$ Liver Transplant (1st year) Liver Transplanted (after 1st year)

„ „

Solid lines for natural history Dashed lines for treatment success 8

4

General Model „ „

Finite decision epochs T = {1, 2, …, N} States: S = {(h,i)} where ‰ ‰

„

Actions: A = {{NT, T, TT} where ‰ ‰

„

h ∈ {1, 2, ..., H}; 1 is healthy; H is death state i ∈ {0,1} or {unaware, aware} of infection NT = do nothing; T = test; TT = test and treat AS are feasible actions for state S, where AS∈A ∀ s∈S

Policy π is set of actions for all times 9

General Model „

rt(s,a) is immediate cost or utility ‰

‰ ‰ ‰

„

pt(s' | s,a) is probability move from s to s’ ‰

‰

„

Immediate cost/utility loss of test or treatment (no future impact) Disease health costs/reduction in utility of disease states Person’s contributions to society if cost is objective Expected cost/utility loss due to secondary infections Can change with time or age due to disease progression or risk behavior Can change when test indicates presence of disease

Minimize total discounted cost or Maximize utility gained ‰ ‰

minπ Eπ { ∑t=1N λt-1 rt(s,πt) }, where λ is discount factor Let utπ(s)=rt(s,πt)+∑j ∈ S λ pt(j |s,πt)ut+1π(j) or total cost/uility onwards 10

5

Structural Results „

Assumption 1 (A1): ‰

„

Advanced disease is more costly or has worse utility

Assumption 2 (A2): ‰

Increasing failure rate „

„

qt((k,i)|s,a)=∑j=kHpt((j,i)|s,a) is nondecreasing in k for every k ∈ {1,…,H}, i ∈ {0,1}, t ∈ {1,…,N-1}, for s ∈ S and for every a ∈ As.

Lemma 1: ‰

‰

Under A1 and A2, for any policy π, utπ((h,i)) is nondecreasing in h, for i ∈ {0,1} and for every t. (Being at a worse health state has a higher cost-to-go)

11

Structural Results: Behavior Change „

Risky behavior affects rewards and probabilities ‰

‰

‰

„

Let pt(s'|s,a), rt(s,a), qt(s'|s,a) and utπ(s) be the parameters for a less risky person. If a person changes his behavior positively, we assume that progression of the disease is slower „ qt((k,i)|s,a) · qt((k,i)|s,a) ∀ k ∈ {1,…,H}, i ∈ {0,1}, t ∈ {1,…,N-1}, for s ∈ S and ∀ a ∈ As

We assume that rt(s,a) · rt(s,a) ∀ periods t since infection of others is lower

Lemma 2: ‰

‰

Under A1 and A2, for any policy π, utπ(s) · utπ(s) for every s ∈ S, and ∀ t. (Behavioral change can reduce your cost-to-go no matter what state you are in) 12

6

Structural Results „

Theorem 1: ‰

‰

Under A1 and A2, and when policy π is used before time t Testing: „

‰

If Eπ[rt(s,T)] · Eπ[rt(s,NT)], then at time t it is better to test than not to test.

Testing & Treating (π uses action NT and π' uses action TT): „

If Eπ' [rt(s,T)]+ λ Eπ' [rt+1(s,π't+1)] · Eπ [rt(s,NT)] +λEπ[rt+1(s,πt+1)], then at time t it is better to test the person (and treat if he is sick) than not to test. 13

Structural Results: Special Case „

Assume there are only two health states (healthy and sick) ‰

„

Modeling short time horizon or disease is not fatal

Theorem 3: ‰

‰

c1 (c2) is cost of testing (treating), q is success probability of the treatment, p is probability of a healthy person not getting the disease at each t, and r1 (r2) is cost per period of being at the healthy (sick) state. If the person is healthy at time 0 and at most one test is allowed, it is beneficial to test and treat the person at time N-t, if „ c1/(1-pN-t) + c2 · λq (r2-r1) [(1-(λp)t)/(1-λp) -r1(λp)t] 14

7

(Negative) Results „

Multiple tests should not necessarily be evenly spaced ‰ ‰

„

Consistent with Diehl et al (2006) Can depend on costs and behavior parameters

Total cost is not necessarily convex in the number of tests

15

Application to Chlamydia „ „

Sexually transmitted disease with mild (or no) symptoms Consider universal testing of women age 20 ‰ ‰ ‰ ‰

„

Probability of infection/year is 2.3% Treatment 93% effective at $15 cost Diagnostic test costs $17.34 Assume healthy cost is $0 and λ = 0.97

Theorem 3 condition holds if yearly cost of being sick > $35.74 ‰

‰ ‰

Two lifetime estimates of cost are $1210 (van Valkengoed et al 2001) and $1524 (Wang et al 2002) Estimated annual cost of disease is > $43.72 Corresponds with CDC recommendation 16

8

Hepatitis C Progression Rx Uninfected

Infected w/out Cirrhosis

Compensated Cirrhosis

Hepatocellular carcinoma

Decompensated Cirrhosis $$ Liver Transplant (1st year) Liver Transplanted (after 1st year)

„ „

Solid lines for natural history Dashed lines for treatment success 17

Application to Hepatitis C: Costs Parameter

Value

References

Cost of screening test (ELISA)

$24.42

Stein et al (2004)

Cost of combination therapy (peginterferon + ribavirin)

$22,896

DMD America (2006)

Discount factor for costs

3%

Lipscomb et al (1996)

Discount factor for QALYs

3%

Singer (2001)

$494 / year

Sullivan et al (2004)

Decompensated cirrhosis

$25,691 / year

Sullivan et al (2004)

Transplantation (1st year)

$312,804 / year

Sullivan et al (2004)

Transplantation (after 1st year)

$30,121 / year

Sullivan et al (2004)

Hepatocellular carcinoma

$16,748 / year

Sullivan et al (2004)

Health State Compensated cirrhosis

18

9

Application to Hepatitis C: Rates Transition Probabilities

Yearly Value

References

Progression Rate to Compensated Cirrhosis with Excessive Alcohol

0.017676

Poynard et al 2001

Progression Rate to Compensated Cirrhosis with Excessive Alcohol

0.007244

Freeman, Wiley

Progression Rate to Compensated Cirrhosis without Excessive Alcohol

0.003622

Poynard et al 2001

Progression Rate to Decompensated Cirrhosis

0.039

Bennett et al (1997)

Progression Rate to HCC from Cirrhosis or decompensated Cirrhosis

0.0268

Degos et al (2000)

Rate of liver transplant from Decompensated Cirrhosis

0.03

Bennett et al (1997)

Death rate from Decompensated Cirrohosis

0.218

Fattovich et al (1997)

Death rate from HCC

0.427

Fattovich et al(1997)

Death rate after Liver transplant first year

0.137

Forman et al (2002)

Bennett et al (1997);

Death rate after Liver transplant after first year

0.052

Forman et al (2002)

Death rates from other causes

(table)

CDC Deaths (2006)

Probability of Treatment Success in non-cirrhotic state

0.54

Manns et al (2001)

Probability of Treatment Success in Compensated Cirrhotic state

0.39

Horoldt et al (2006)

We assume disease awareness reduces alcohol to < 50 grams/day

19

Application to Hepatitis C: Rates Age

Death Rates

Under 1 year

0.007

1-4

0.000315

5-9

0.000147

10-14

0.000191

15-17

0.000664

18-19

0.000664

20-24

0.000964

25-29

0.000952

30-34

0.001113

35-39

0.001591

40-44

0.002413

45-49

0.003592

50-54

0.005178

55-59

0.007556

60-64

0.011827

65-69

0.017937

70-74

0.027783

75-79

0.043634

80-84

0.069762

85 and over

0.145933

Risk Group [estimated age ranges & reference] Overall population [13-55, CDC]

Annual Incidence

References

0.0004

CDC Hepatitis Survey (2006)

Injection drug Users (IDU) [15-50, HHS]

0.014

CDC Viral Hepatitis C (2006)

Commercial sex Workers [15-45, Brewer et al 2000]

0.0012

National Network (2005)

We assume: 1) probability of secondary infection equal to own probability of infection 2) disease awareness reduces secondary infection by 50% (Singer and Younossi 2001)

20

10

Application to Hepatitis C: More Data % of VirusGenotype 1

60%

Treatment Success rate of Genotype 1

29%

Hornberger(2006) Hornberger(2006)

Treatment Success rate non-Genotype 1

62%

Hornberger(2006)

ELISA Test, False Negative

0.014

Singer(2001)

ELISA Test, False Positive

0.009

Singer(2001)

Cost of Infecting Others

$15,525

calculated by model

QALY of Infecting Others

-1.1

calculated by model

% of overall population that drink alcohol excessively

4.90%

BFRSS Chong (2003)

QALYs

Singer (2001)

Uninfected

1

1

Infected Without Cirrhosis

0.96

0.79

Compensated cirrhosis

0.8

0.8

Decompensated cirrhosis

0.56

0.6

Transplantation (1st year)

0.8

0.73

Transplantation (after 1st year)

0.95

0.73

Hepatocellular carcinoma

0.25

QALY during Treatment

0.72 0.93 (Bennett)

21

Results with Single Test (Overall)

22

11

Results with Single Test

23

Results for Multiple Tests

Additional Dsicounted Utility

Up to 5 Tests per lifetime for the Overall Population with QALY values from Singer and Alcohol Progression value 0.0072 100 80

Base Case No Alcohol Reduction Treat G2 Only

60 40 20 0 0

1

2

3

4

5

Number of Tests per Lifetime

24

12

Impact of Alcohol Behavior Effectiveness 35000 30000 25000 20000 15000 10000 5000 0

AFP 0.0072 AFP 0.017

75 % 50 % 25 % 0% Alcohol Alcohol Alcohol Alcohol Reduction Reduction Reduction Reduction

Additional Discounted Utility

Base Case

Additional Discounted Utility by % Alcohol Effectiveness for the CSW Population with QALY values from Singer 3000 2500 2000 1500 1000 500 0

APF .0072 AFP .071

Base Case

75 % Alcohol Reduction

50 % Alcohol Reduction

25 % Alcohol Reduction

0 % Alcohol Reduction

25

Single Test with Cost/QALY Single Test per Lifetime for overall population with QALY values from Singer and Alcohol Progression value .0072 Additional Discounted Cost

Additional Discounted Utility

Additional Discounted Utility by % Alcohol Effectiveness for the IDU Population with QALY values from Singer

120 38 34 37 4140 3936 3533 474645444342 32 504948 30 31 535251 28 29 565554 26 27 57 24 25 58 23 59 22 60 21 61 20 6362 19 6564 18 66 67 17 6968 16 7170 15 72 73 74 14 75 76 77 13 79 78 80

100 80 60 40 20

Single Test $50K/QALY line

0 0

0.001

0.002

0.003

0.004

Additional Discounted QALYs

26

13

Multiple Tests with Cost/QALY (IDU)

27

Multiple Tests with Cost/QALY (Overall)

28

14

3 Tests with Cost/QALY (Overall) 3 Test + Treat (overall population) Singer QALY values and Alcohol Progression=.0072

Additional Discounted Cost

200 180 160

16 19,19, 18023, 16 23,14 14 21, 25, 17,0180 19, 25, 14 1614 27, 0 14 15,15, 17, 18 17, 20 0 140 00 0 19,16 20 000 00 10 12 13,13, 1816 0 21,17, 25, 18 20 012 27, 31, 0 2029, 10 23, 10 0 17,0 19, 22 22 0 031, 088 25, 012 0 0 29, 21,033, 22 21, 8 13, 120 23, 0 20 22 0 0 27, 0 00025, 22 0 0 00 35, 0 831, 009, 27, 0 2 18 22 6 0 17, 8 33, 16 0 31, 6 0 37,00 826031, 29,20 22 27, 24 23, 26 17,15, 2608 37,21, 0 12 35,0 629, 24 19, 39,21, 8 28 16 27, 31, 26 22 06 35, 0 37, 0 0 060 13, 26 39, 10 3, 22 27, 0320 33, 44 37,035, 16 39, 12 19, 30 0 35, 22 41, 10 39, 16 43, 6 41,39, 12 4 41, 4 14 41, 29, 13, 30 35, 22 15, 32 13, 6 17, 4 43, 4 41, 18 45, 8 23, 2 0

0

140 120

47, 8 13, 4 51, 2

0

43, 18

Evenly Spaced 3 Tests Convex Hull of 3 Tests $50K/QALY line

49, 4 47, 14 49, 1047, 16 49, 14

100

53, 6 13, 2

51, 12

53, 12 55, 6 55, 12 57, 4 57, 10

80

59, 6 61, 6 63, 4

60

65, 2 67, 2 69, 2 71, 4

40 20 0 0

0.002

0.004

0.006

Additional Discounted QALYs 29

When to Test (Summary) „

Testing Recommendations ‰

‰

„ „

High-risk groups should be tested (multiple times) over a wide range of ages Testing is also (barely) cost-effective for the general population

Dynamic screening is appropriate in some cases Further analytical analysis could be useful for designing screening policies for various diseases 30

15

Where to Test?? „ „

Statistical analysis of Hepatitis C National Health And Nutrition Examination Survey (NHANES) combines survey and exam for sample in the US ‰ ‰

N = 8,369 subjects age 20 – 59 Prevalence is 2.3% with 48.8% undiagnosed!

31

Factors for Hepatitis C Coefficient

Odds Ratio

95% CI

Gender

0.82

0.48-1.40

Black, Non-Hispanic

2.07

1.30-3.32

Hispanic

0.74

0.42-1.33

Poverty (Income < 200%FPL)

1.81

1.07-3.05 2.08-6.34

Race

„

≥ 20 lifetime sex partners

3.63

Elevated liver enzymes

8.82

5.29-14.74

Ever used IV drugs

70.78

37.61-132.49

Had transfusion prior to 1992

2.57

1.35-4.89

Fair or poor oral health

3.09

1.88-5.07

Fair or poor general health

1.11

0.58-2.11

Logistic regression with reference group white non-Hispanic, female, family income > 200% of FPL, less than 20 sex partners in lifetime, never used IV drugs, no blood transfusion prior to 1992, reporting good to excellent oral health and general health 32

16

Factors for Hepatitis C „ „

Applied same model to undiagnosed Hepatitis C Significant variables ‰ ‰

Poor general health (lower than reference group) Poor oral health!! (higher than reference group) „

„

„

If you have poor oral health, you are 4 times more likely to have Hep C If you have poor oral health, the probability of undiagnosed disease is 2 times higher

This may imply a “where” 33

Next Steps „

„

Screening for poor oral health may imply a location (e.g., dental offices) Ongoing work to assess cost-effectiveness and value of early detection

34

17

Mathematical Modeling for Health Policy „

„

„

Analysis of bathhouse closures for impact on HIV transmission Design of food (vaccine) distribution network for response to a pandemic Comparison of healthcare access through optimally located community clinics to coverage by Medicaid

35

Questions?

Julie Swann, PhD Georgia Tech [email protected]

36

18