An AI-based quantitative home-use framework for assessing fertility and identifying novel hormone trends by recording urine hormones

Testing with IFM and repredicition from calibration curve

Calibration curve for each batch of Inito Fertility test strips was generated using standard solutions prepared in spiked urine. The ODs obtained from standard solutions were plotted against concentration and this plot was used to re-predict concentrations in all further experiments. For testing, the test strips were dipping in the urine samples for 15 seconds. The strips were then inserted into the Inito Fertility Monitor attached to mobile and values of E3G, PdG and LH were obtained along with the fertility ratings for each day. In the Inito Fertility test strip, E3G and PdG are measured in a competitive ELISA assay format where the intensity of respective test lines decreases with increasing concentration and LH is measured in a sandwich ELISA format where the intensity of the test line increases with increasing concentration.

Sample preparation for characterization of Inito Fertility Monitor

Samples used for precision studies, linearity of re-prediction of concentration and cross-reactivity studies were prepared by spiking male urine samples with target concentrations of the metabolites. The male urine samples were tested with ELISA to confirm negligible concentrations of the respective metabolites. Purified metabolites for all studies were obtained from Sigma-Aldrich chemicals as described in supplementary table 1.

Testing with ELISA kits

The same user samples were also used in an ELISA to measure the exact concentration. E3G and PdG were measured using the Arbor Estrone-3-Glucuronide EIA kit (K036-H5) and the Arbor Pregnanediol-3-Glucuronide EIA kit (K037-H5) ELISA kits. Urinary LH was measured using the DRG LH (urine) ELISA kit (EIA-1290). For all runs, solutions of fixed concentration (provided along with the respective kits) were used to generate the standard curve and the concentration of metabolites in urine samples were calculated from the standard curve generated. All samples were measured in triplicates and the average value was considered for comparison.

Study participants

Two groups of women were recruited for the study. The first group consisted of women from whom the first urine sample of the day was collected over one menstrual cycle. The second groups consisted of women who had expressed previous interest over the internet to try the Inito Fertility Monitor. Women in the second group were provided the IFM and the Inito Fertility Test strips to be tested at home. The study protocol was approved by the appropriate institutional review boards (ISRCTN15534557). Informed consent forms were obtained from all volunteers. Women aged between 21-45 years of age were recruited for the study. Women were excluded if:

1. They were using hormonal contraceptives, including oral, emergency oral, implants, patches, transdermal injections, vaginal ring and progesterone intrauterine systems (IUS)

2. Were on infertility medications or hormone replacement therapy containing hCG or LH

3. Were consuming clomiphene citrate or other ovulation induction drugs

4. Have recently been pregnant, miscarried or breastfeeding

Reasons for data exclusion were irregularities in testing leading to insufficient data points for any conclusion and failure to meet the selection criteria. The characteristics of those who were considered for the study are summarized in Table 1.

Study design

Urine samples collected from women in the first group were frozen, transported to the testing site and tested with the IFM on the same day. Freeze-thaw has been previously shown to have no effect on the concentration of urinary hormones^12,13. For the second group, IFM and Inito Fertility test strips were sent to participants to perform the test at home. Standard instructions for use were provided with the IFM box and the test strips. All participants had the option to contact the manufacturer’s helpline number for any additional information concerning the IFM. Women began using the IFM at the start of the menstrual cycle after admission into the study. Participants were instructed to take tests on specific cycle days when instructed by the application. Hormone data from all participants was collected via the Inito application. Although women were recruited for one cycle each, participants were excluded if they did not take tests on the required days The study design is summarized in Figure 2.

• Download figure
• Open in new tab

Figure 2.

Flow of participants through the study

Inito fertility monitor accurately predicts the concentration of urine metabolites and is specific to the metabolites of interest

Accuracy of the Inito Fertility test strips was assessed by re-predicting the concentrations of E3G, PdG and LH in standard solutions prepared in male urine not containing any of the metabolites. Six spiked solutions containing all the three metabolites were prepared from respective stock solutions as per Supplementary Table 2. Each solution was tested with five different Inito fertility test strips and the recovery percentage was calculated. We found that the average recovery percentage across five tests was 100.16% for E3G, 104.63% for PdG and 104.74% for LH. The observed concentration obtained from IFM was also highly correlated to the expected concentration in the spiked male urine (Fig.3a-c) implying a high accuracy in predicting the concentrations.

• Download figure
• Open in new tab

Figure 3.

Correlation between observed concentration as obtained from IFM and expected concentrations of E3G (a), PdG (b) and LH (c). Linear regression (blue) along with the 95% CI (dotted black line) for the comparison between re-predicted hormone values from urine samples obtained using IFM and ELISA: E3G (d), PdG (e) and LH (f).

Since Inito Fertility test strips are manufactured in a batch wise fashion, we wanted to ensure that the accuracy is maintained across lots. Therefore, ten strips from different lots were tested with four different concentrations of E3G, PdG and LH and the coefficient of variation (CV) was calculated. All three assays were found to have a CV of less than 10% across different lots (Supplementary Table 3), indicating that the recovery percentage does not vary significantly from test strip to test strip. Therefore, IFM is an accurate device to measure urinary hormones and is consistent from test to test.

Next, we wanted to analyse the specificity of Inito fertility test strips towards the metabolites of interest. We identified the components of human urine that could potentially interfere with the test and spiked negative male urine with these compounds. The concentration of each metabolite was selected to encompass normal and pathological conditions. We tested four strips from different batches with the spiked male urine and found that the test strips did not cross-react with any urine metabolite at even the highest possible physiological condition (Supplementary Table 4). This suggests that the Inito fertility test strips are specific to the metabolites of interest only.

Urinary hormone concentrations estimated by IFM and ELISA are strongly correlated

Enzyme linked-immunosorbent assay (ELISA) is a standard laboratory procedure for measuring concentrations of chemical compounds accurately. Therefore, we wanted to compare the efficacy of estimation by Inito with respect to ELISA. For this, we chose urine samples from three random user cycles of the 100 volunteer cycles. This accounted for a total of 80 samples that were tested with both Inito and ELISA and the correlation was calculated. We found that results obtained from ELISA and Inito are highly similar (r²=0.96 for E3G, r²=0.99 for PdG, and r²=0.99 for LH) and are linearly correlated (Fig.3d-f) implying that measurement by Inito is almost as accurate as measurement by a standard laboratory technique.

Incidence of premature E3G surge and multi-phasic LH surge maybe higher in women than expected

Our study design allowed us to monitor urinary E3G, LH and PdG on all days of the follicular phase and the luteal phase including the day of ovulation. Therefore, we decided to observe the occurrence of aberrant E3G and LH patterns in the volunteer menstrual cycles. Specifically, we investigated the prevalence of premature E3G surge and multi-phasic LH surge (multiple LH peaks). We observed that among the cycles which displayed LH rise and ovulation (confirmed using urinary PdG), 39 cycles recorded multiple surges of LH before the actual peak that caused ovulation and 16 cycles recorded a single dominant LH surge. 39 cycles were anovulatory and had no LH surge. However, 6 cycles were anovulatory despite a single surge of LH, a phenomenon previously described as luteinized anovulation. While the occurrence of multi-phasic LH is coincident with previous studies^14,15, the percentage of cycles with a single LH surge leading to ovulation may actually be lower than what had been previously assumed. With respect to E3G, we observed that 20 cycles showed a significantly high E3G in the early part of the cycle and before the expected E3G surge (right before LH surge) the occurrence of which is coincident, albeit slightly higher than previously reported studies¹⁵.

Monitoring continuous rise of urinary PdG after LH surge provides early confirmation of ovulation

Serum progesterone (P4) concentration is widely considered as a confirmatory marker for ovulation. Typically, a mid-luteal phase value of >3ng/mL has been used to confirm ovulation. Recently, it has been shown that mid-luteal phase measurement of urinary PdG (a threshold value of >5ug/mL) has a good correlation with serum P4 behaviour and can also be used to confirm ovulation^16,17. However, both these methods require the users/patients to wait for almost 7 days after the LH surge.

For our analysis, to classify a menstrual cycle as ovulatory or anovulatory, we selected the same urinary PdG threshold values as previously discussed and this was regarded as the actual positive or negative value for further evaluation of the novel criteria. The baseline urinary PdG was determined before the LH surge as the average of the PdG levels and compared with the PdG levels 2-3 days after the LH surge. We found that the distribution of fold change in ovulatory cycles was substantially different from the fold change observed in the anovulatory cycles during the first three days after the LH peak (Fig.4a and b). From this observation, we inquired if measuring a particular fold rise within 3 days after the LH peak could be used as a criteria to predict ovulation earlier. We decided to judge this criteria by performing an ROC analysis. We found that a fold change of 2.75x within 3 days could accurately distinguish ovulatory from anovulatory cycles. The criteria had a specificity and sensitivity of 100% in detecting ovulation and was significant in differentiating the ovulatory from anovulatory cycles (p<0.0001, AUC:0.981, Fig.4c and Table 2a). In addition, we applied this criterion to the set of users who used the IFM at home to predict ovulation. We found that the new criteria could differentiate the 18 ovulatory cycles from the 4 anovulatory cycles with a 100% precision (Table 2b). Our findings show that constant urinary PdG monitoring can provide earlier ovulation confirmation instead of a single PdG/P4 measurement in the mid-luteal phase.

• Download figure
• Open in new tab

Figure 4.

a. Representative hormone chart for multi-phasic LH surges with each event of LH surge marked (black arrow), b. Representative graph showing PdG rise before the LH surge (blue arrow) and the gradual rise in PdG after LH peak (black arrow) for confirming ovulation as opposed to a single test mid-luteal phase (red arrow), c. Comparison of fold rise in PdG levels within 3 days after LH surge in ovulatory versus anovulatory cycles (****p<0.0001 and d. ROC analysis to evaluate the accuracy of the proposed criteria to predict ovulatory and anovulatory cycles.

Surge in luteinizing hormone is preceded by a distinct rise in the urinary PdG

A surge in estradiol levels prior to the LH surge is observed in canonical menstrual cycles. Progesterone, however, is thought to be the baseline during the follicular phase (Fig.1a). Our study design allowed us to monitor urinary PdG prior to LH surge as well. Furthermore, the sensitivity of the Inito Fertility test strips allowed the smallest changes in the urinary PdG to be visualised. We observed that urinary PdG increased in Group I prior to the LH surge by an average of 3.16+/-0.7 fold in 52/55 cycles from the baseline stage (in cycles where there was a prominent LH surge). Just 2 cycles illustrated this phenomenon in the 6 anovulatory cycles of Group I, where there was an LH surge. In Group II, we found the frequency of this phenomenon in 43/52 cycles by re-evaluating this observation. Although we remain unsure of the source of this urinary PdG, the phenomenon appears to be highly conserved during menstrual cycles.