Triaging pregnancies of unknown location: the performance of protocols based on single serum progesterone or repeated serum hCG levels

How does a protocol based on a single serum progesterone measurement perform as a triage tool in women with pregnancy of unknown location (PUL) in comparison to protocols based on serial hCG measurement? Triage based on the logistic regression model M4 (using initial hCG and hCG ratio (48 h/0 h)) classifies the majority of PUL into low and high risk groups, in contrast to a progesterone protocol based on a serum level threshold of 10 nmol/l. Low progesterone has been shown to identify failing pregnancies and those at low risk of complications. A prediction model (M4) based on the initial hCG and the hCG ratio at 0 and 48 h can successfully classify PUL into low and high risk groups. A multi-centre diagnostic accuracy study of 1271 women was performed retrospectively on data from women at St. Georges Hospital (SGH, London, UK) between February 2005 and 2006, Queen Charlottes Chelsea Hospital (QCCH, London, UK) between April 2009 and August 2012, and the Royal Prince Alfred Hospital (RPAH, Sydney, Australia) between February 2008 and October 2011. The end-points were the final observed outcome for each pregnancy as a failed PUL (low risk), intrauterine pregnancy (IUP, low risk), or ectopic pregnancy (EP, high risk), and any interventions or complications for EP during the follow-up period. Complete data were available for initial progesterone, 0/48 h hCG and final outcome in 431 of 534 women (81) at SGH, 396/585 (68) at QCCH and 96/152 (63) at RPAH. Missing values were handled using multiple imputation. Three diagnostic approaches were used to classify PUL as high risk: a range of serum progesterone levels were evaluated (10, 16 and 20 nmol/l) for the progesterone protocol, risk of EP given by the M4 model 5 for the M4-based protocol, and hCG ratio was between 0.87 and 1.66 for hCG cut-offs as previously published. Results were analysed using random intercept models or stratified analysis to account for variability between centres. The progesterone protocol based on levels of 10 nmol/l classified 24 (95 confidence interval 2028) of failed PUL, 95 (9297) of IUP and 76 (6783) of EP as high risk. The M4 protocol classified 14 (1117) of failed PUL, 37 (3143) of IUP and 84 (7690) of EP as high risk. The hCG ratio cut-offs classified 10 (812) of failed PUL, 15 (1120) of IUP and 63 (5371) of EP as high risk. Using complete cases only, 67 of EP treated with methotrexate (n 48) and 89 surgically managed (n 37) were correctly classified by the progesterone protocol, 96 and 81 by M4 protocol and 75 and 65 by hCG ratio cut offs, respectively. Data were incomplete for 103 (19), 189 (32) and 56 (37) patients at SGH, QCCH and RPAH, respectively; however, we are reassured by the minimal differences seen between the results of complete cases and those following imputation of missing values. The variation in the inclusion criteria between the three centres is also a potential limitation of this study; however, it reflects real clinical practice. Furthermore, the hCG ratio cut-offs were not originally developed to optimize triage. The results show that serum progesterone is less efficient for triage than serial hCG measurements assessed using the M4 model, the striking difference being serum progesterone places nearly all IUP in the high-risk category. A two-step strategy combining single-visit and two-visit approaches should be investigated. Funding was from Research Foundation-Flanders (FWO). There are no competing interests.