Cost-Utility Analysis of Deep Learning and Trained Human Graders for Diabetic Retinopathy Screening in a Nationwide Program

Srisubat, Attasit; Kittrongsiri, Kankamon; Sangroongruangsri, Sermsiri; Khemvaranan, Chalida; Shreibati, Jacqueline Baras; Ching, Jack; Hernandez, John; Tiwari, Richa; Hersch, Fred; Liu, Yun; Hanutsaha, Prut; Ruamviboonsuk, Varis; Turongkaravee, Saowalak; Raman, Rajiv; Ruamviboonsuk, Paisan

doi:10.1007/s40123-023-00688-y

Table 1 Input parameters used in the base case analysis

From: Cost-Utility Analysis of Deep Learning and Trained Human Graders for Diabetic Retinopathy Screening in a Nationwide Program

Parameters	Distribution	Mean	SE	Source
Transitional probability
No DR to non-STDR	Beta	0.0524	0.0010	[11]
No DR to STDR	Beta	0.0086	0.0010	[11]
Non-STDR to STDR	Beta	0.0734	0.0010	[11]
No DR to BB	Beta	0.0016	0.0002	[13, 14]
Non-STDR to BB	Beta	0.0040	0.0004	[13, 14]
STDR to BB	Beta	0.0080	0.0006	[13, 14]
Efficacy of treatment
Relative risk of DR progression when patients were treated	Beta	0.4900	0.0689	[15]
Prevalence
Prevalence of diabetic patients without DR	Dirichlet	0.7236	0.7236	[11]
Prevalence of diabetic patients with non-STDR	Dirichlet	0.1537	0.1537	[11]
Prevalence of diabetic patients with STDR	Dirichlet	0.1227	0.1227	[11]
Proportion of DME among STDR	Beta	0.9247	0.0925	[11]
Relative risk of mortality in diabetic patients	Gamma	1.8500	0.1850	[16]
Relative risk of mortality in diabetic patients with non-STDR	Gamma	1.1300	0.1990	[17]
Relative risk of mortality in diabetic patients with STDR	Gamma	2.2600	0.6633	[17]
Relative risk of mortality in diabetic patients with blindness	Gamma	2.2600	0.6633	Assumed: using the risk of patients with STDR as stated in text
Efficacy of screening
Sensitivity of screening by HG	Beta	73.72%	1.66%	[11]
Specificity of screening by HG	Beta	98.59%	0.17%	[11]
Sensitivity of screening by DL	Beta	95.03%	0.82%	[11]
Specificity of screening by DL	Beta	97.97%	0.20%	[11]
Uptake screening, HG	Beta	50.00%	5.00%	Primary data from Health Data Center (HDC) of Ministry of Public Health in the Fiscal year 2020
Compliance to follow-up confirmation after positive screening result, HG	Beta	60.00%	8.00%	Data from random sampling
Uptake screening, DL	Beta	50.00%	5.00%	Assumed: equal to HG based on model purpose
Compliance to follow-up confirmation after positive screening result, DL	Beta	60.00%	8.00%	Assumed: equal to HG based on model purpose
Direct medical cost
Cost of screening
Unit cost of image grading by DL per person (THB)	Gamma	32	3	Set for this study
Unit cost of image grading by HG per person (THB)*	Gamma	56.11	5.6	[18]
Unit cost of image grading (confirmation) by retina specialist per person (THB)	Gamma	112.22	11.22	[18]
Unit cost of colour fundus image capture (both eyes) (THB)	Gamma	194	19	Primary data
Unit cost of visual acuity measurement (THB)	Gamma	40	4	[19]
Unit cost of mydriasis (THB)	Gamma	73	7	[19]
Unit cost of OPD service per visit in a primary facility (THB)	Gamma	109	11	[19]
Unit cost of OPD service per visit in a tertiary facility (THB)	Gamma	309.50	30.95	[19]
Unit cost of IPD service per day of admission (THB)	Gamma	1329	133	[19]
Cost of treatment
Cost of PRP (THB)	Gamma	7000	700	Primary data
Cost of PPV in 2020 (2-day admission or one-day surgery) (THB)	Gamma	30,000	2968	Primary data
Unit cost of bevacizumab per one injection (THB)**	Gamma	769	107	Primary data
Unit cost of an intravitreal administration and topical antibiotic (THB)	Gamma	1719	172	Primary data
Unit cost of OCT (THB)	Gamma	700	70	Primary data
Frequency of treatment using PPV	Gamma	1.00	0.10	[20]
Proportion of the patients with PDR who undergo PPV	Beta	0.33	0.03	[20]
Frequency of treatment using PRP***	Gamma	2.00	0.20	[21]
Frequency of OP visit (per year) for PRP in STDR without DME: 1st year	Gamma	2.00	0.20	Assumed: according to screening and treatment frequencies
2nd–5th year	Gamma	1.00	0.10
Subsequent year	Gamma	1.00	0.10
Frequency of IVB (per year) for DME; 1st year	Gamma	10.00	1.00	[21, 22]
Frequency of IVB (per year) for DME; 2nd–5th year	Gamma	1.75	0.18	[21, 22]
Frequency of OP visit (per year) for DME; 1st year	Gamma	12.00	1.20	Assumed: according to screening and treatment frequencies
Frequency of OP visit (per year) for DME; 2nd–5th year	Gamma	5.25	0.53
Frequency of OP visit (per year) for DME; Subsequent year	Gamma	2.00	0.20
Direct non-medical cost
Cost of travel for OPD per visit, primary facility (THB)	Gamma	59	12	[19]
Cost of travel for OPD per visit, tertiary facility (THB)	Gamma	156	31	[19]
Cost of food for OPD per visit, primary facility (THB)	Gamma	15	3	[19]
Cost of food for OPD per visit, tertiary facility (THB)	Gamma	57	11	[19]
Patient time spent for OP visit at primary facility (min)	Gamma	69	14	[19]
Patient travel time spent for OP visit at primary facility (min)	Gamma	18	4	[19]
Patient travel time spent for OP visit at tertiary facility (min)	Gamma	60	12	[19]
Caregiver time spent for care to patient with BB (hours/week)	Gamma	28	6	[13]
Gross National Income per capita (THB/year)	Gamma	240,586	48,117	[23]
Utilities
Utility of patients with no DR	Beta	0.748	0.026	[24]
Utility of patients with non-STDR (mild to moderate NPDR)	Beta	0.752	0.037	[24]
Utility of patients with STDR without DME	Beta	0.628	0.052	[24]
Utility of patients with DME****	Beta	0.628	0.052	Assumed
Utility of DR patients with blindness	Beta	0.355	0.128	[24]

The relative risks of mortality among patients in the different health states were combined with age-specific all-cause mortality derived from the Thai population’s life table data [25]. Direct medical costs of treatment for STDR and DME were estimated by assuming 75% of patients have the bilateral DR. This cost of image grading by DL per patient should cover practice expense of DL in Thailand considering the large population of patients screened in the national program. All incurred costs were converted to 2020 values using the consumer price index for Thailand and were converted to US$ using the exchange rate as of 1 July 2021 of 32.02 THB per US$ [26]
BB bilateral blindness, CVD cardiovascular disease, DL deep learning, DR diabetic retinopathy, HG human graders, IPD inpatient department, IVB intravitreal bevacizumab, OCT optical coherence tomography, OP outpatient, OPD outpatient department, PDR proliferative diabetic retinopathy, PPV Pars Plana vitrectomy, PRP panretinal photocoagulation, STDR sight-threatening diabetic retinopathy, THB Thai baht, NHSO National Health Security Office
*Adjusted from the monthly salary of trained human graders at 26,000 THB; **according to Thai National Health Security Office reimbursement policy, 1 vial of bevacizumab 100 mg/4 ml is divided into 30 doses; ***panretinal photocoagulation was divided to be performed twice in an eye as a standard practice; ****patients with STDR with or without DME were assumed to have the same utility

Back to article page

Search

Navigation