Clinical and economic evaluation of a proteomic biomarker preterm birth risk predictor: cost-effectiveness modeling of prenatal interventions applied to predicted higher-risk pregnancies within a large and diverse cohort

Figures & data

Figure 1. Schematic overview of the ACCORDANT study. ACCORDANT subjects (N = 847) were routed to three study arms: standard care (SoC) control; risk predictor plus case management intervention (RP-CM); or risk predictor plus multimodal management intervention (case management with pharmacological treatment) (RP-MM). Those in the RP-CM and RP-MM arms were stratified into higher-risk and lower-risk groups based on their proteomic biomarker risk predictor scores, using a previously validated threshold. Subjects predicted to be at higher preterm birth risk were modeled to receive case management or multimodal management. Lower-risk subjects in the RP-CM and RP-MM arms received standard care. The gestational ages at birth of higher-risk subjects were shifted according to published intervention efficacies. Associated outcomes were calculated for the treated groups and for subjects with missing or truncated data in both arms. Outcomes were compared between active arms and the control arm, as a whole and between the most severely affected 10% of each arm by outcome.

Figure 2. Shifts in gestational ages at birth applied to subjects who were predicted to be at higher risk for preterm birth and subsequently modeled to receive case management (increased outreach, preterm education and specialist care) or multimodal management (case management with pharmaceutical treatment). Shifts were based on published intervention efficacies.

Table 1. Summary of ACCORDANT assumptions.

Input	Type	Center	Spread	Distribution	Reference
Size of population	Demographic	As observed	As observed	N/A	Markenson^14
Composition of population	Demographic	As observed	As observed	N/A	Markenson^14
Test performance	Clinical	As observed	As observed	N/A	Saade^9
Untreated GA at birth	Clinical	As observed	As observed	N/A	Markenson^14
Untreated neonatal LOS	Clinical	As observed	As observed	N/A	Markenson^14
Untreated maternal LOS	Clinical	As observed	As observed	N/A	Markenson^14
Untreated neonatal morbidity and mortality	Clinical	As observed	As observed	N/A	Markenson^14
Multimodal management intervention	Clinical	max_m = 9.5 weeks mid_m = 30.75 weeks rate_m = −1/2.3^†	max_SD = 3.09 weeks mid_SD = 30.13 weeks rate_SD = −1/2.9^†	Three-parameter sigmoid	Branch^12
Case management intervention	Clinical	max_m = 5.25 weeks mid_m = 30.75 weeks rate_m = −1/2.1^†	max_SD = 3.09 weeks mid_SD = 30.13 weeks rate_SD = −1/2.9^†	Three-parameter sigmoid	Stankaitis^16 Newman^17 Newnham^18
Treated neonatal LOS for GA < 35 weeks (MIX)	Clinical	GA-dependent as published; 80% of mean applied to gamma, 20% to lognormal deviates	GA-dependent as published; 60% of variance applied to gamma, 40% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	MIX^22
Treated neonatal LOS for GA ≥ 35 weeks (MIX)	Clinical	GA-dependent as published; 10% of mean applied to gamma, 90% to lognormal deviates	GA-dependent as published; 90% of variance applied to gamma, 10% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	MIX^22
Treated neonatal LOS for GA < 35 weeks (COM)	Clinical	GA-dependent as published; 85% of mean applied to gamma, 15% to lognormal deviates	GA-dependent as published; 60% of variance applied to gamma, 40% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	COM^23
Treated neonatal LOS for GA ≥ 35 weeks (COM)	Clinical	GA-dependent as published; 15% or mean applied to gamma, 85% to lognormal deviates	GA-dependent as published; 90% of variance applied to gamma, 10% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	COM^23
Treated maternal LOS for GA < 35 weeks	Clinical	GA-dependent as published; 20% of mean applied to gamma, 80% to lognormal deviates	GA-dependent as published; 60% of variance applied to gamma, 40% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	MIX^22
Treated maternal LOS for GA ≥ 35 weeks	Clinical	GA-dependent as published; 20% of mean applied to gamma, 80% to lognormal deviates	GA-dependent as published; 80% of variance applied to gamma, 20% to lognormal deviates	Method of moments: replicating parametric and non-parametric summary statistics	MIX^22
Treated neonatal morbidity and mortality	Clinical	See Supplemental Table 1		Multinomial	Markenson^14
Treated or untreated neonatal LOS for cases of perinatal mortality	Clinical	One day longer than maximum observed	0	N/A	Branch^12
Neonate cost for cases of perinatal mortality	Economic	$89,069	$0	Point estimate: average observed	MIX^22
Neonate cost	Economic	GA-dependent cost per day as published applied to neonatal LOS	As for neonatal LOS	Method of moments: blend of gamma and lognormal	MIX^22
Neonate cost	Economic	GA-dependent cost per day as published applied to neonatal LOS	As for neonatal LOS	Method of moments: blend of gamma and lognormal	COM^23
Neonate cost adjustment	Economic	MIX^22 (12/2017) to TREETOP (4/2019): 1.026; COM^23 (2016) to TREETOP (4/2019): 1.049	n/a	N/A	FRED^24
Progesterone cost	Economic	$20	$0	Point estimate	Fried^25 Cohen^26 Cahill^27 Kowalski^28
Case management cost	Economic	$53	$0	Point estimate	Estimated from CMS physician fee schedule for PREVENT^12 protocol care

^† max_m and max_SD are the means and standard deviations for the upper asymptotes; mid_m and mid_SD are the means and standard deviations for the inflection points; and rate_m and rate_SD are the means and standard deviations for the slopes of the three parameter sigmoid functions.

Abbreviations. CMS, Centers for Medicare & Medicaid Services; COM, U.S. commercial payer dataset; GA, gestational age; LOS, hospital length of stay; MIX, U.S. state-based dataset with a mix of insurance payers.

Figure 3. Impact of test-and-treat strategies on neonatal length of hospital stay. The population assessed was either (A) all subjects or (B) those in the top 10% with respect to length of stay. *p < .05, determined using Cox proportional hazards regression. **p < .001.

Abbreviations. COM, U.S. commercial payer dataset; MIX, U.S. state-based dataset with a mix of insurance payers.

Figure 4. Impact of test-and-treat strategies on neonatal length of hospital stay (A,B) and neonatal costs (C,D) among self-identified Black and Hispanic individuals. The population assessed was either all such subjects (A,C) or those in the top 10% with respect to length of stay or costs (B,D). Abbreviations. COM, U.S. commercial payer dataset; MIX, U.S. state-based dataset with a mix of insurance payers.

Figure 5. Impact of test-and-treat strategies on neonatal costs. The population assessed was either (A) all subjects or (B) those in the top 10% with respect to costs. *p < .05, determined using bootstrap intervals. Abbreviations. COM, U.S. commercial payer dataset; MIX, U.S. state-based dataset with a mix of insurance payers; ns, non-significant.

Table 2. Impact of test-and-treat strategies on neonatal morbidity and mortality.

Arm	Level 1 (mild) (%)	Level 2 (moderate) (%)	Level 3 (severe) (%)	Level 4 (death) (%)	Proportion moderate+ (%)	p-Value^† (Level 1 versus > 1)	p-Value^† (Levels 1–2 versus > 2)	p-Value^† (Levels 1–3 versus 4)	% Reduction in moderate +
Risk Predictor + Case Management (RP-CM)	93.3	4.7	1.9	0.1	6.7	0.025	0.14	0.19	29%
Risk Predictor + Multimodal Management (RP-MM)	94.5	4.1	1.4	0.0	5.5	0.0015	0.035	0.19	41%
Standard of Care (SoC)	90.6	6.5	2.5	0.5	9.4	–	–	–	–

Comparisons were made between the control arm and the two active arms of the proportion of subjects at or below versus above each elevated level of the neonatal morbidity and mortality index. The percent reduction in levels 2 or higher is shown for each active arm. Significant p-values are indicated in boldface type.

^† Fisher’s Exact test; significance, p < 0.05.

Markenson GR, Saade GR, Laurent LC, et al. Performance of a proteomic preterm delivery predictor in a large independent prospective cohort. Am J Obstet Gynecol MFM. 2020;2(3):100140.

 PubMed Web of Science ®Google Scholar

Saade GR, Boggess KA, Sullivan SA, et al. Development and validation of a spontaneous preterm delivery predictor in asymptomatic women. Am J Obstet Gynecol. 2016 May;214(5):633.e1-633–e24.

 Web of Science ®Google Scholar

Branch DW, VanBuren JM, Porter TF, et al. Prediction and prevention of preterm birth: a prospective, randomized intervention trial. Am J Perinatol. 2021. DOI:10.1055/s-0041-1732339

 Web of Science ®Google Scholar

Stankaitis JA, Brill HR, Walker DM. Reduction in neonatal intensive care unit admission rates in a medicaid managed care program. Am J Manag Care. 2005;11(3):166–172.

 PubMed Web of Science ®Google Scholar

Newman RB, Sullivan SA, Menard MK, et al. South Carolina partners for preterm birth prevention: a regional perinatal initiative for the reduction of premature birth in a Medicaid population. Am J Obstet Gynecol. 2008;199(4):393.e1–393.e8.

 PubMed Web of Science ®Google Scholar

Newnham JP, White SW, Meharry S, et al. Reducing preterm birth by a statewide multifaceted program: an implementation study. Am J Obstet Gynecol. 2017;216(5):434–442.

 PubMed Web of Science ®Google Scholar

Phibbs CS, Schmitt SK, Cooper M, et al. Birth hospitalization costs and days of care for mothers and neonates in California, 2009–2011. J Pediatr. 2019;204:118–125 e14.

 PubMed Web of Science ®Google Scholar

Beam AL, Fried I, Palmer N, et al. Estimates of healthcare spending for preterm and low-birthweight infants in a commercially insured population: 2008–2016. J Perinatol. 2020;40(7):1091–1099.

 PubMed Web of Science ®Google Scholar

Federal Reserve Bank of St. Louis. FRED economic data: producer Price Index by Industry: selected Health Care Industries (PCUASHCASHC) 2022. [cited 2022 Feb 26]. Available from: https://fred.stlouisfed.org/series/PCUASHCASHC

 Google Scholar

Fried I, Beam AL, Kohane IS, et al. Utilization, cost, and outcome of branded vs compounded 17-alpha hydroxyprogesterone caproate in prevention of preterm birth. JAMA Intern Med. 2017;177(11):1689–1690.

 PubMed Web of Science ®Google Scholar

Cohen AW, Copel JA, Macones GA, et al. Unjustified increase in cost of care resulting from U.S. Food and Drug Administration approval of Makena (17alpha-hydroxyprogesterone caproate). Obstet Gynecol. 2011;117(6):1408–1412.

 PubMed Web of Science ®Google Scholar

Cahill AG, Odibo AO, Caughey AB, et al. Universal cervical length screening and treatment with vaginal progesterone to prevent preterm birth: a decision and economic analysis. Am J Obstet Gynecol. 2010;202(6):548 e1–548 e8.

 Web of Science ®Google Scholar

Kowalski JT. A guide to the cost of progesterone for prevention of preterm labor. Proc Obstet Gynecol. 2011;1(3):9. Available from: http://ir.uiowa.edu/pog/

 Google Scholar

Data availability statement

Data supporting the results presented here can be requested by emailing the corresponding author. Data will not be made available publicly or in any format that may violate a subject’s right to privacy.