Combining reproductive outcomes predictors and automated estrus alerts recorded during the voluntary waiting period identified subgroups of cows with different reproductive performance potential

The objective was to compare differences in reproductive performance for dairy cows grouped based on the combination of data for predictors available during the prepartum period and before the end of the VWP, automated estrus alerts (AEA) during the VWP, and the combination of both factors. In a cohort study, data for AEA and potential predictors of the percentage of cows that receive insemination at detected estrus (AIE) and pregnancies per AI (P/AI) for first service, and the percentage of cows pregnant by 150 DIM (P150) were collected from −21 to 49 DIM for lactating Holstein cows (n = 886). The association between each reproductive outcome with calving season (cool, warm), calving-related events (yes, no), genomic daughter pregnancy rate (gDPR; high, medium, low), days in the close-up pen (ideal, not ideal), health disorder events (yes, no), rumination time (high or low CV prepartum and high or low increase rate postpartum), and milk yield (MY) by 49 DIM (high, medium, low) were evaluated in uni-variable and multivariable logistic regression models. Individual predictors (health disorders, gDPR, and MY) associated with the 3 reproductive outcomes in all models were used to group cows based on risk factors (RF; yes, n = 535 or no, n = 351) for poor reproductive performance. Specifically, cows were included in the RF group if any of the following conditions were met: the cow was in the high MY group, had low gDPR, or had at least one health disorder recorded. Cows were grouped into estrus groups during the VWP based on records of AEA (E-VWP, n = 476 or NE-VWP, n = 410). Finally, based on the combination of levels of AEA and RF cows were grouped into an estrus and no RF (E-NoRF, n = 217), no estrus and RF (NE-RF, n = 276), no estrus and no RF (NE-NoRF, n = 134), and estrus and RF (E-RF, n = 259) groups. Cows received AIE up to 31 d after the end of the VWP, and if did not receive AIE, received timed AI after an Ovsynch plus progesterone protocol. Logistic and Cox proportional hazard regression compared differences in reproductive outcomes for different grouping strategies. The NoRF (AIE: 76 .9 %; P/AI: 53 .1 %; P150: 84 .5 %) and E-VWP (AIE: 86 .8 %; P/AI: 44 .8 %; P150: 82 .3 %) groups had more cows AIE, P/AI, and P150 than the RF (AIE: 64 .5 %; P/AI: 34 .9 %; P150: 72 .9 %) and NE-VWP (AIE: 50 .0 %; P/AI: 38 .9 %; P150: 72 .1 %) groups, respectively. When both factors were combined, the largest and most consistent differences were between the E-NoRF (AIE: 91 .3 %; P/ AI: 58 .7 %; P150: 88 .5 %) and NE-RF groups (AIE: 47 .3 %; P/AI: 35 .8 %; P150: 69 .5 %). Compared with the whole population of cows or cows grouped based on a single factor, the E-NoRF and NE-RF groups had the largest and most consistent differences with the whole cow cohort. The E-NoRF and NE-RF group also had statistically significant differences of a large magnitude when compared with the remaining cow cohort after removal of the respective group. We conclude that combining data for AEA during the VWP with other predictors of reproductive performance could be used to identify groups of cows with larger differences in expected reproductive performance than if AEA and the predictors are used alone.


INTRODUCTION
Reproductive performance is paramount for the profitability of dairy farms as it directly influences herd milk production efficiency, the herd replacement dynamics, and cost of production (Vries, 2006;Cabrera, 2010;Giordano et al., 2012).As the reproductive efficiency of dairy herds continues to improve and herd management needs to evolve, novel approaches are necessary to make the most efficient use of available resources and optimize outputs.In this regard, targeted reproductive management (TRM) is an approach to herd management that consists of identifying subgroups of cows within a herd that share features associated with reproductive performance (Giordano et al., 2022).These subgroups of cows with expected similar performance receive tailored management strategies that enhance herd performance, profitability, or management to a greater extent than if the whole herd is managed with the same program or receive the same intervention.Examples of tailored programs include different duration of the voluntary waiting period (VWP) (Gobikrushanth et al., 2014;Stangaferro et al., 2018a;Edvardsson Rasmussen et al., 2023), favoring AI at detected estrus (AIE) or TAI (Fricke et al., 2014;Rial et al. ., 2022;Gonzalez et al., 2023), hormonal interventions before or after AI (Bisinotto et al., 2015;Giordano et al., 2015;Zolini et al., 2019), and targeted use of different type of semen (Kaniyamattam et al., 2018;Berry, 2021).
A first and critical step in the development of TRM strategies is identifying and characterizing the predictors of reproductive outcomes needed to create subgroups of cows with sufficient variability in performance potential as to justify tailored management.Predictor data must be collected and available in advance of reproductive management decision-making points and explain variability of cow responses to specific management practices.Therefore, data collected and available during the VWP of lactating dairy cows is of interest and potential value for use in TRM (Giordano et al., 2022;Rial et al. ., 2022;Gonzalez et al., 2023).
Automated estrus alerts (AEA) recorded during the VWP have been recently explored and used as a predictor for TRM because of the strong association between expression of estrus in early lactation and key reproductive performance outcomes.Cows with at least one, or 2 or more AEA within the first 45 to 60 d after parturition were more likely to receive AIE, had more pregnancies per AI (P/AI) to first service, and had greater pregnancy rate and more cows were pregnant by mid-to-late lactation (Borchardt et al., 2021;Rial et al., 2022;Bretzinger et al., 2023).Based on these associations, TRM strategies aimed at optimizing first service management programs using AEA recorded during the VWP have been recently developed and evaluated (Fricke et al., 2014;Rial et al., 2022;Gonzalez et al., 2023).These programs had some positive effects including reduced days to first service, greater pregnancy rate in early lactation, and optimized use of hormonal interventions.Nevertheless, the full potential of this approach might not have yet been realized because of the limited magnitude of the differences in performance between groups of cows with or without AEA during the VWP.Novel strategies to identify subgroups of cows with larger variability in reproductive potential than through AEA alone could increase the performance, management, or profitability benefits of TRM because greater differences in biological potential of the targeted groups might increase the value of tailored interventions (Giordano et al., 2022;Sitko et al., 2023).
Grouping cows based on levels of multiple factors might help identify subgroups with greater differences in performance than when a single factors is used.In this regard, several associations between cow, herd management, and environmental features data available immediately before and early after parturition have been associated with reproductive performance and could be combined with AEA to create groups of cows for TRM.For instance, health disorders within the first 50 DIM were associated with fewer pregnancies per AI (P/AI), reduced pregnancy rates, and increased pregnancy losses (Ribeiro et al., 2016;Carvalho et al., 2019).A poor transition from late gestation into lactation due to nutritional and metabolic imbalances associated with or caused by suboptimal herd management or environmental conditions has been associated with impaired fertility in multiple studies (Ribeiro et al., 2016;Carvalho et al., 2019;Pinedo et al., 2020).High milk yield for cows in the same herd or managed under the same conditions has been linked to poorer reproductive performance including fewer P/AI and decreased pregnancy rate in some studies (Santos et al., 2009;Stangaferro et al., 2018a).Associations between genetic merit for fertility traits based on genomic predictions with reproductive performance have also been widely reported (Chebel and Veronese, 2020;Lima et al., 2020;Sitko et al., 2023).Cows with superior genomic predicted transmitting ability for reproductive traits had more P/AI, fewer pregnancy losses, and greater pregnancy rates than cows of inferior genetic merit for fertility (Chebel and Veronese, 2020;Lima et al., 2020;Sitko et al., 2023).The number and diversity of predictors of reproductive potential available during the VWP that could be combined with AEA is large.Nevertheless, the effect of combining these multiple predictors with AEA recorded during the VWP to generate subgroups of cows with different reproductive performance potential for TRM has not been previously investigated in dairy cattle.
Thus, our primary hypothesis was that grouping cows based on the combination of cow features, performance, and events data (herein "predictors") recorded during the prepartum and VWP periods with AEA recorded during the VWP would identify groups of cows with greater differences in reproductive performance than if grouping was based on AEA alone, combinations of predictors alone, or the whole cow cohort (i.e., no grouping).To this end, our first objective was to compare the reproductive performance of dairy cows grouped based on the combination of data for predic-Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE tors available during the prepartum period and VWP, cows grouped based on AEA recorded during the VWP, and the combination of both so that subgroups of cows with the largest and most consistent differences in performance could be identified.Thereafter, our second objective was to evaluate differences in reproductive performance outcomes for cows grouped based on different combinations of factors positively or negatively associated with reproductive performance and either the whole cow cohort or the cohort of cows left after removal of these groups.

MATERIALS AND METHODS
All procedures performed with animals were approved by the Animal Care and Use Committee of Cornell University (protocol #2017-0024).

Cows and Herd Management
Data from a randomized control trial conducted at a commercial dairy farm located in Long Prairie, Minnesota from May 2020 to November 2021 were used in this cohort study.The farm was selected for the experiment because the automated estrus monitoring system required (Smartbow, Zoetis) for data collection was already installed and functional, DairyComp305 (ValleyAg Software) was used for dairy herd management, and the farm management team agreed to allow the researchers to conduct the study procedures.The farm housed an average of 1,259 (range: 1,231 to 1,287) milking and 166 (range: 139 to 203) dry cows during the study period.During the prepartum period, cows were housed in the same freestall barn in nulliparous and parous cow pens with deep, sand-bedded stalls, and fans and sprinklers above the feed lane for heat abatement.Within ~3 to 12 h after parturition, cows were transported to a lactating cow barn where primiparous and multiparous cows were commingled in a single post-fresh pen until ~21 DIM.Cows were milked 3 × /d at ~8 h intervals in a double 18 parallel parlor equipped with individual stall milk meters (Dematron 75, GEA) for capturing milk weights at every milking.The projected 305-d milk yield estimated by Dairy-Comp305 for cows that calved during the study period was 11,340 kg (range: 7,030 to 18,048 kg) and average daily milk yield was 37.2 kg/d (range: 34.9 kg/d to 39.4 kg/d).Cows had ad libitum access to water and a TMR delivered 2 × /d with feed pushed up every hour.Diets provided during the prepartum, and lactating periods were the same for primiparous and multiparous cows.

Study Design and Data Collection
This cohort study used data from a randomized controlled trial designed to evaluate the effect of AEA during the VWP as a tool for TRM (Rial et al., 2022).All cows (n = 1,035) that calved and initiated a lactation during the study period were eligible for enrollment, except for cows assigned to the "do not breed" group by farm personnel or cows that left the herd before 21 DIM.Data were collected from 21 d before parturition through 150 DIM.
Potential predictors available and used in data analysis were season of calving (cool vs warm), calvingrelated events (CRE), genomically enhanced daughter pregnancy rate (gDPR), days in the close-up pen (DCU), health disorder (HD) events diagnosed after calving, rumination time (RT) and milk yield (MY; Figure 1).Data for season of calving, gDPR, DCU, CRE, and HD diagnosed after calving until 21 DIM were collected once because a single value was possible during the study period.Health disorder diagnosis was conducted following disorder criteria definition and standard operating procedures described previously (Rial et al., 2022).Data for gDPR available in the dairy herd management software were used as an indicator of genetic merit for fertility.Cows were genotyped using a commercially available test (Clarifide, Zoetis, Parsippany, N).The median gDPR value for cows of the same parity group was imputed for 91 cows without gDPR data.Data for daily RT were collected from 21 d before to 20 d after calving and daily milk yield data were collected up to 49 DIM.
Automated estrus alerts were recorded from 21 to 49 DIM.All AEA recorded on the day of or the day after a pen move event (1.9%; 31/1,568) were considered false positive alerts.This restriction was imposed because of the high occurrence of false positive AEA when cows are moved between pens at commercial farms (Rial and Giordano, personal observation).Based on previous observations at dairy farms using automated estrus detection systems and rigorous health monitoring programs that include daily manipulation of cows in the fresh pen, concerns arose about potential false positive alerts due to intensive handling of fresh cows.Distinguishing true AEA from false positive alerts caused by cow handling and pen movements is difficult unless estrus is confirmed.As the study lacked a method to confirm estrus in cows with AEA from 7 to 20 DIM, these AEA were ignored.Moreover, out of the cows that had an AEA before 21 DIM, most (177/212) had another alert from 21 to 49 DIM and thus, were included in the E-VWP group.
At 50 DIM all cows were categorized into the estrus (E-VWP) or the no-estrus (NE-VWP) VWP group

Reproductive Management
Regardless of the estrus status during the VWP, cows were randomly assigned to a non-TRM or a TRM program that prioritized submission of cows to first service through AIE.After the end of a 49 d VWP regardless of the estrus status during the VWP, cows in the non-TRM program (P-AIE; n = 517) were eligible to receive first service through AIE for 24 ± 3 d after the end of the VWP.For cows that did not receive AIE, TAI occurred at 83 ± 3 DIM after synchronization of ovulation with an Ovsynch protocol with progesterone supplementation and 2 PGF 2α treatments (P4-Ov; GnRH plus intravaginal P4-releasing device, 7 d later PGF 2α and P4-releasing device removal, 24 h later PGF 2α , 32 h later GnRH, and 16-20 h later TAI) as described in Rial et al. (2022).After the end of a 49 d VWP, cows in the TRM program (TP-AIE; n = 518) with estrus detected during the VWP were eligible for AIE for 31 ± 3 d whereas cows with no estrus detected during the VWP were eligible for AIE for 17 ± 3 d after the end of the VWP.Cows that did not receive AIE received the same synchronization of ovulation protocol (i.e., P4-Ov) than cows in the P-AIE treatment.Thus, cows in the TP-AIE treatment received TAI at 90 ± 3 were detected in estrus during the VWP or 74 ± 3 DIM if were not detected in estrus during the VWP.
For second and greater AI services, all cows were managed with the same program consisting of AIE before pregnancy diagnosis through AEA as described.
Cows not re-inseminated at detected estrus underwent pregnancy diagnosis and if not pregnant, were submitted to an Ovsynch protocol with 2 PGF 2α treatments (GnRH, 7 d later PGF 2α , 24 h later PGF 2α , 32 h later GnRH, and 16 h to 20 h later TAI).
Inseminations were conducted 2 × /d (0530 to 0930 h and 1430 to 1730 h) by 2 farm technicians while cows were restrained in headlocks in the feeding lane.All AIE services were conducted if cows had an AEA as defined previously within the eligible period for each group.Specifically, every morning at 0500 h, farm personnel manually created a list of cows with AEA occurring ≥8 h prior.The same process was conducted at 1400 h.Cows with AEA that started <8 h before the creation of the morning list were added to the afternoon insemination list.Timed AI was conducted from 0530 h to 0930 h in cows that completed synchronization of ovulation protocols according to their treatments.
Pregnancy diagnosis and confirmation of pregnancy were conducted by the farm veterinarian by transrectal palpation of the uterus at 39 ± 3 d and 74 ± 3 d after AI, respectively.Pregnancy loss was estimated from the day of the initial pregnancy diagnosis until confirmation of pregnancy.

Data Collection and Processing for Cow Grouping
Combination of predictors data for creation of risk factor groups.Data for putative predictors available on the day of calving were DCU, pre-calving RT, gDPR, season of calving, and CRE.Summary statistics for data for the predictors is presented in Supplementary Table 1.Data for these predictors were used to categorize cows as follows: ideal (≥14 and ≤28) or not ideal (<14 or >28) DCU, cool (calvings from January through April) or warm (calvings from May through September) season of calving, presence or absence of CRE (at least one of the following: twins, initiation of lactation after an abortion [expulsion of a fetus and placental membranes after 220 and before 260 d of gestation], stillbirth, or dystocia), and low, medium, or high gDPR for cows grouped by parity.Days in close-up pen to classify cows as ideal or not ideal were defined based on the distribution of DCU for cows in the current study (Supplementary Table 1).The lower tertile upper limit for gDPR was −0.2 and 0.0 and the highest tertile lower limit was 0.9 and 0.9 for primiparous and multiparous cows, respectively.Use of RT data to classify cows during the prepartum period is described with the post-calving period.
Data for potential predictors available postpartum were HD events, MY, and RT (Figure 1).Cows were categorized based on data for these predictors as follows: none (No) or one or more (Yes) HD (at least one recorded event of metritis, mastitis, displaced abomasum, indigestion, ketosis, lameness, or pneumonia) events recorded up to 20 DIM.Cows were classified in low, medium, or high MY groups based on tertiles of accumulated milk yield up to 49 DIM within parity group (primiparous: lower tertile upper limit = 1,402 kg and highest tertile lower limit = 1,606 kg, multiparous: lower tertile upper limit = 2,018 kg and highest tertile lower limit = 2,300 kg).Using data for RT prepartum and postpartum, the coefficient of variation (CV) of daily RT for the last 21 d prepartum and the rate of increase of RT after calving were calculated.The CV was obtained using PROC MEANS of SAS.Cows were then divided into high and low CV groups based on the median CV for the prepartum period for primiparous (CV prepartum RT median = 5.80) and multiparous (CV prepartum RT median = 5.50) cows.The rate of increase for daily RT from parturition until the DIM at which the RT plateau (RTIR) was observed in this data set was calculated.The difference between the value at the RT plateau (average RT from 10 to 13 DIM) and the value at baseline (average RT from 2 to 4 DIM) was divided by the baseline value.The RT plateau and baseline were calculated based on 3 daily values to reduce the effect of daily variability in RT.Based on individual cow values, tertiles were created for primiparous (lowest tertile upper limit = 10%; highest tertile lower limit = 25%) and multiparous (lowest tertile upper limit = 17%; highest tertile lower limit = 40%) cows.
The number and percentage of cows in each level for the individual predictors in presented in Supplementary Table 2.

Statistical Analyses and Creation of Groups.
The primary reproductive performance outcomes of interest for this study were the percentage of cows AIE for first service based on AEA, P/AI to first service, and percentage of cows pregnant by 150 DIM.Secondary outcomes of interest were time to first service and time to pregnancy by 150 DIM to complement the analysis of percentage of cows AIE and percentage of cows pregnant by 150 DIM.No a priori sample size was conducted as we used a convenient sample of data available from a previous conducted randomized controlled experiment (Rial et al., 2022).
All data analyses were conducted using SAS software (version 9.4, SAS Institute Inc.).
After cows were assigned a value for levels (e.g., yes or no, high, medium, or low) of individual predictors, univariable logistic regression models fitting a binomial distribution with PROC GLIMMIX were created to evaluate the association between each predictor and binary outcomes of interest.Subsequently, all predictors with P < 0.20 in univariable models were offered to multivariable logistic regression models fitting a binomial distribution with PROC GLIMMIX.Parity group (i.e., primiparous and multiparous) was forced in all multivariable models whereas type of semen was offered as confounder to P/AI and pregnancy by 150 DIM models only.Backward elimination of variables with P > 0.10 was adopted to select final multivariable models.All predictors initially offered to univariable or multivariable models but removed because of lack of an association with outcomes were offered as covariates in the remaining analyses.
Grouping based on risk factors for poorer reproductive performance.Putative predictor variables associated (P ≤ 0.05) with the 3 binary outcomes of interest in multivariable models were selected for creation of risk factor (RF) groups for poor reproductive performance.Cows with levels of individual predictors associated with reduced reproductive performance (i.e., reduced percentage of cows AIE, fewer P/AI, and fewer cows pregnant at 150 DIM) were considered RF for poorer performance and therefore cows with these levels were included in the RF group.Specifically, cows were included in the RF group if any of the following conditions were met: the cow was in the high MY group, had low gDPR, or had at least one health disorder recorded.The remaining cows were included in the NoRF group.
Grouping based on AEA during the VWP.Cows were also grouped based on AEA recorded dur-Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE ing the VWP independently of RF groups.Cows with at least one AEA recorded from 21 to 49 DIM were included in the estrus VWP (E-VWP) group, whereas cows with no AEA were included in the no estrus VWP group (NE-VWP).Grouping based on AEA during the VWP was based on previous evidence of differences in reproductive performance between cows with and without AEA during the VWP (Borchardt et al., 2021;Rial et al., 2022).
Grouping based on the combination of RF and AEA.After cows were included in the RF or NoRF and the E-VWP or NE-VWP group, another grouping variable was created based on the combination of RF and AEA group as follows: AEA during the VWP and no RF recorded (E-NoRF), no AEA during the VWP and no RF recorded (NE-NoRF), AEA during the VWP and RF recorded (E-RF), and no AEA during the VWP and RF recorded (NE-RF).
Analyses of reproductive performance based on grouping strategies.Multivariable logistic regression models with PROC GLIMMIX fitting a binomial distribution were created to evaluate the association between each outcome of interest and the RF group only (RF vs NoRF), estrus group only (E-VWP vs NE-VWP), and the groups formed based on the combination of RF and AEA (E-NoRF, NE-NoRF, E-RF, and NE-RF).Time to event outcomes (i.e., days to first service and days to pregnancy for 150 DIM) for different grouping strategies were analyzed using Cox's proportional hazard regression with PROC PHREG.The assumption of proportionality of the hazards was visually assessed by plotting the logarithm of the negative logarithm of the survival probability by the logarithm of days at risk using PROC LIFETEST.Variables not used for grouping cows were offered to all models as covariates.These analysis aimed to evaluate if there was a positive association between not presenting RF and presenting AEA with the outcomes of interest.The analysis for the combined groups was aimed to evaluate putative stronger associations with the outcomes of interest when cows had either the most favorable or unfavorable combinations of levels of RF and AEA than when presented only the favorable or unfavorable levels of the factors RF and AEA.A manual backward stepwise elimination procedure was used to remove covariates with P > 0.10.Parity was forced as a fixed effect in all models.Models for P/AI to first service were also offered type of semen (sexed or beef) used and the reproductive management program used to submit cows to first service (P-AIE or TP-AIE) as fixed effect covariates.
Comparison of magnitude of differences in reproductive performance for different grouping strategies.To test the hypothesis that cow grouping strategies would generate differences of different magni-tude for outcomes of interest, logistic regression models created with the response variable modeled using the events over trials syntax of PROC LOGISTIC of SAS were used to evaluate the percentage of cows AIE, P/ AI, and percentage of cows pregnant by 150 DIM when cows were not grouped based on RF or AEA (i.e., whole cow cohort), when cows were grouped based only on AEA or RF, and the combination of both.For this analysis, the denominator represented the total number of cows within a group and the numerator included only cows in which the outcome of interest was observed (insemination at detected estrus for AIE and pregnancy for P/AI and percentage of cows pregnant by 150 DIM).Because there are 2 levels within with RF, AEA, and grouping combining AEA and RF, separate analyses were conducted to compare outcomes among groups of cows that had characteristics either favorably (i.e., E-VWP, NoRF, and E-NoRF) or unfavorably (i.e., NE-VWP, RF, and NE-RF) associated with reproductive performance.In both cases, groups compared in analyses were all cows in the data set (i.e., no grouping) versus the respective groups.Finally, to demonstrate the potential practical application of selecting a single group of cows for targeted management within a herd, analyses were conducted to compare differences in the 3 primary outcomes of interests between the E-NoRF or the NE-RF group and the remaining cow cohort after removing cows in these groups.
Except for the univariable models used to explore associations between individual predictors and outcomes of interest (i.e., used P < 0.20), for all analyses of reproductive performance based on different grouping strategies, all explanatory variables were considered significant if P ≤ 0.05 whereas 0.05 < P ≤ 0.10 were considered tendencies.When appropriate, the least significant difference (LSD) post hoc mean separation test was used to determine differences between least squares means (LSM).All percentages reported are LSM and the 95% CI were generated using the LSMEANS statement of GLIMMIX.

Study Population
Out of 1,035 cows enrolled, 149 were removed from analyses because were sold (n = 33) or died (n = 7) before the first service, had missing values for RT (n = 91), or were not in compliance with experimental procedures (n = 18).Out of the 886 cows used for analyses, 60.4% (535/886) presented at least one characteristic negatively associated with reproductive outcomes of interest and therefore were included in the RF group.Specifically, 60.4% of the cows presented HD, were in Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE the high MY group, or were in the low gDPR group.As cows could have presented more than one RF, 6.5% had HD and were in the high MY group, 9.3% had HD and were in the low gDPR group, 20.7% were in the high MY and low gDPR group, and 2.2% had HD, were in the high MY group and were in the low gDPR group (percentage of cows with each predictor within parity are presented in Supplementary Table 2).The remaining 39.6% (351/886) of the cows were included in the NoRF group.On the other hand, 53.7% (476/886) of the cows presented at least one AEA during the VWP and were included in the E-VWP group while the remaining 46.3% (410/886) of the cows were included in the NE-VWP group.Based on the combination of AEA during the VWP and RF for reduced reproductive performance, 24.5% (217/886) of the cows were included in the E-NoRF group, 15.1% (134/886) in the NE-NoRF group, 29.3% (259/886) in the E-RF group, and 31.1% (276/886) in the NE-RF group.

Creation of Risk Factors Groups: Associations between Predictors and Reproductive Outcomes based on Univariable and Multivariable Models
Associations between outcomes of interest and potential predictors from univariable logistic regression analyses are presented in Table 1 and results from multivariable models are presented in Table 2.

Percentage of Cows AIE
Variables selected from univariable models (P < 0.20) were the number of days in the close-up pen, HD group, RTIR group, MY group, and gDPR group.No association was observed for the other potential predictors.
Based on the multivariable model, a greater (P < 0.05) percentage of cows were AIE for the group without HD, the low than the medium and high MY groups, and the high than the medium and low gDPR groups.There was no association between DCU (P = 0.37) and RTIR (P = 0.11) with the percentage of cows AIE.

Pregnancies per AI
Variables selected from univariable models (P < 0.20) were calving season, CRE group, HD group, RTIR group, MY group, and gDPR group.The other potential predictors were not associated with P/AI.
Based on the multivariable model, more (P < 0.05) P/AI were observed for cows without HD, cows in the medium than the high MY group (no differences for the low and the other MY groups), and cows in the high than the low gDPR group (no differences for the medium and the other gDPR groups).Season of insemination (P = 0.07), CRE (P = 0.07), and RTIR (P = 0.16) were not associated with P/AI.

Cows Pregnant by 150 DIM
Variables selected from univariable models (P < 0.20) were days in the close-up pen, HD group, RTIR group, MY group, and gDPR group.No association was observed for the other potential predictors.
Based on the multivariable model, the percentage of cows pregnant at 150 DIM was greater (P < 0.05) for the group with no HD, the medium MY group, and the medium and high gDPR groups.Also, more cows were pregnant at 150 DIM the group with an ideal number of days in the close-up pen.There was no association between RTIR (P = 0.19) and cows pregnant by 150 DIM.

Differences in Reproductive Performance Based on Different Grouping Strategies
Percentage of cows AIE.The percentage of cows AIE was greater for the NoRF than the RF group (P < 0.001; Figure 2A) and the hazard ratio for time to first service was greater for the NoRF (P < 0.001; HR = 1.39; 95% CI: 1.21-1.59)than the RF group.
A greater percentage of cows received AIE in the E-VWP than the NE-VWP group (P < 0.001; Figure 2B) and the hazard ratio for time to first AI was greater for the E-VWP than the NE-VWP group (P < 0.001; HR = 1.3; 95% CI: 1.2-1.5).
Pregnancies per AI.The No-RF group had more P/AI than the RF group (Figure 2C) and the E-VWP group tended to have more P/AI than the NE-VWP group (Figure 2D).More P/AI were observed for the E-NoRF and NE-NoRF groups than for the E-RF and NE-RF groups (Figure 4A).
Cows Pregnant and Time to Pregnancy by 150 DIM.The percentage of cows pregnant was greater for the No-RF than the RF group (P < 0.001; Figure 2E) and greater for the E-VWP than the NE-VWP group (P = 0.04; Figure 2F).The E-NoRF had a greater (P < 0.001) percentage of cows pregnant than the E-RF and NE-RF groups but did not differ from the NE-NoRF group (Figure 4B).The NE-RF group had the smallest percentage of cows pregnant compared with the other groups.Moreover, the E-NoRF group had greater (P < 0.001) hazard of pregnancy (HR = 1.48; 95% CI: 1.17-1.87)than the NE-NoRF group, the E-RF group (HR = 1.65; 95% CI: 1.35-2.02),and the NE-RF group (HR = 2.26; 95% CI: 1.83-2.78;Figure 5).Likewise, the NE-NoRF group had greater hazard of pregnancy (HR = 1.53; 95% CI: 1.21-1.94)than the NE-RF group and the E-RF group had greater hazard of pregnancy (HR = 1.37; 95% CI: 1.12-1.67)than the NE-RF group.No difference was observed between NE-NoRF and E-RF groups.

Grouping based on favorable characteristics
The percentage of cows AIE was affected by the grouping strategy (P < 0.001) because more cows received AIE (Figure 6A) and the odds of receiving AIE were greater for cows in the E-NoRF group than for cows in the entire cow cohort (OR = 4.3, 95% CI: 2.7-6.9;P < 0.001), and the E-VWP group (OR = 2.9, 95% CI: 1.7-4.9;P = 0.004), and tended to differ with the NoRF group (OR = 1.6, 95% CI: 0.9-2.6;P = 0.08), Thus, the magnitude of the difference in the percentage of cows AIE between the entire cow cohort and groups with favorable characteristics was greatest with the E-NoRF (23.1%) and E-VWP (18.6%) groups, and smallest for the NoRF group (8.7%).
Pregnancies per AI differed among grouping strategies (P < 0.001) because more cows were pregnant to first service (Figure 6B) and the odds of becoming pregnant at first service were greater for cows in the E-NoRF group than for cows in the entire cohort (OR = 1.7, 95% CI: 1.2-2.2;P < 0.001) and the E-VWP group (OR = 1.5, 95% CI: 1.1-2.1;P = 0.01).There was no difference between cows in the E-NoRF group and the NoRF group (OR = 1.1, 95% CI: 0.8-1.6;P = 0.43).The magnitude of the difference in P/AI between the entire cow population and groups with favorable characteristics was greatest with the E-NoRF (15.8%) group and NoRF (10.2%) group, whereas there was no difference with the E-VWP (1.9%) group.
The percentage of cows pregnant by 150 DIM differed among grouping strategies (P = 0.002) because more cows were pregnant (Figure 6C) and the odds of becoming pregnant by 150 DIM were greater for the E-NoRF group than for the entire cow population (OR = 1.9, 95% CI: 1.3-3.0;P = 0.002) but the E-NoRF group did not differ from the NoRF (OR = 1.2, 95% CI: 0.8-2.0;P = 0.37) and the E-VWP group (OR = 1.4,95% CI: 0.9-2.2;P = 0.14).Thus, the magnitude of the difference in percent pregnant by 150 DIM between the entire cow population and groups with favorable characteristics was greatest with the E-NoRF group (10.2%) than for the NoRF (6.2%) and E-VWP (4.0%) which were intermediate.

Grouping based on unfavorable characteristics
The percentage of cows AIE was affected by the grouping strategy (P < 0.001; Figure 6D) because fewer cows received AIE and the odds of receiving AIE were lower for cows in the NE-RF group than for cows in the entire cow population (OR = 0.4, 95% CI: 0.3-0.5;P < 0.01), and the RF group (OR = 0.5, 95% CI: 0.4-0.7;P = 0.01), but no difference was observed when compared with cows in the NE-VWP group (OR = 0.9, 95% CI: 0.7-1.2;P = 0.55).Thus, the magnitude of the difference in percentage of cows AIE between the entire cow population with groups that had unfavorable characteristics was greatest with the NE-RF (20.9%) and NE-VWP (18.2%) groups, whereas there was no difference with the RF group (3.7%).
Pregnancies per AI tended to differ among grouping strategies (P = 0.06; Figure 6E) because fewer cows were pregnant at first service and the odds of becoming pregnant at first service were lower for cows in the NE-RF group than for cows in the entire population (OR = 0.7, 95% CI: 0.6-0.9;P = 0.03) but did not differ from cows in the NE-VWP (OR = 0.8, 95% CI: 0.6-1.2;P = 0.82) or the RF group (OR = 0.9, 95% CI: 0.7-1.3;P = 0.33).No differences were observed in the magnitude of the difference in P/AI between the entire cow population and groups with unfavorable characteristics.
The percentage of cows pregnant by 150 DIM differed among grouping strategies (P = 0.006; Figure 6F) because fewer cows were pregnant and the odds of becoming pregnant by 150 DIM were lower for the NE-RF group than for the entire cow population (OR = 0.6, 95% CI: 0.4-0.8;P < 0.01) but did not differ from the RF (OR = 0.7, 95% CI: 0.6-1.1;P = 0.11) and the NE-VWP group (OR = 0.8, 95% CI: 0.6-1.2;P = 0.27).Thus, the magnitude of the difference in the percentage of cows pregnant by 150 DIM between the entire cow population and groups with favorable characteristics was lowest for the NE-RF group (8.8%) than for the NE-VWP (6.2%) and RF (5.4%) groups which were intermediate.

Performance of cows with most favorable or unfavorable characteristics versus the remaining cohort
The percentage of cows that received AIE, P/ AI, and percentage of cows pregnant by 150 DIM were   greater (P < 0.01) for the E-NoRF group than for the rest of the cow cohort (Figure 7A).On the other hand, the percentage of cows that received AIE, P/AI, and percentage of cows pregnant by 150 DIM was reduced (P < 0.01) for the NE-RF group than for the rest of the cow cohort (Figure 7B).

DISCUSSION
Enhancing dairy herd reproductive performance and management through implementation of TRM strategies depends upon the type and magnitude of the improvements in performance and management outcomes for subgroups of cows within a herd (Giordano et al., 2022).Identifying cows with the most extreme and consistent differences in performance potential is needed before targeted interventions, therapies, or management strategies are implemented.Results from this study supported the primary hypothesis that combining dairy cow data readily available during the VWP would enable creating cow groups with large and consistent differences in reproductive performance outcomes.Here, we combined data from cow features such as productivity, genetic potential, health disorders events, and AEA collected before cows were eligible for first service.Compared with grouping cows based on levels of RF for poor reproductive performance or AEA alone, groups with larger differences in performance were created based on the combination of both.Notably, compared with the groups created based on RF or AEA alone (Figures 3 to 5), the whole herd with no grouping (Figure 6), or the remaining cohort of cows after removing the groups with both favorable (i.e., E-NoRF) or unfavorable (i.e., NE-RF) levels of the classification factors (Figure 7), the largest and most consistent differences for percentage of cows AIE, P/AI, and cows pregnant by 150 DIM were observed for the groups created based on the combination of RF and AEA data.Time to event data were also in line with observations for binary outcomes.For instance, the difference in the percentage of cows inseminated at detected estrus was smaller when grouping cows based only on AEA [37 percentage points (p.p.)] than when cows were grouped based on the combination of data for AEA and RF (44 p.p.).Similarly, larger differences were observed for P/AI and cows pregnant by 150 DIM.For example, the difference for P/AI was ~6 p.p. vs ~23 p.p. and the difference for cows pregnant at 150 DIM was ~10 p.p. vs ~19 p.p. when cows were grouped based only on AEA or the combination of AEA and RF levels, respectively.Collectively, these results demonstrated that the combination of data for more than one predictor can identify groups of cows more suitable for TRM strategies due to their substantial differences in reproductive performance potential.
The performance of individuals or groups of cows is the result of complex interactions between several biological, management, and environmental factors (Santos et al., 2009;Ribeiro et al., 2016;Stangaferro et al., 2018b;Manriquez et al., 2021) that individually or jointly affect reproductive outcomes in variable manners and magnitude.This was reflected through variability in the observed magnitude of the associations evaluated.For example, the difference in cows AIE was more than double between the group with or without AEA compared with the difference observed for cows with or without RF factors for poor reproductive performance.Conversely, the difference in P/AI at first service was 3-fold greater for cows grouped based on RF levels than cows grouped based the occurrence of AEA.At least partially, this can be explained by the existence of different biological drivers or magnitude of the effect of such drivers on expression of estrus and fertility.For example, the ability of cows to express estrus and ovulate after calving might be more closely related to the ability of cows to express estrus after the end of the VWP, which is directly linked to the percentage of cows AIE (Borchardt et al., 2021;Rial et al., 2022).Conversely, some of the key biological drivers of fertility such as uterine health and oocyte quality, which are more closely linked to P/AI, might be influenced more and closely associated with biological and management factors used for creating the RF groups such as health disorders and genetic merit for fertility (Ribeiro et al., 2016;Bruinje et al., 2023;Sitko et al., 2023).
On the other hand, data suggested that the relatively smaller differences for percent pregnant by 150 DIM for the E-VWP vs NE-VWP or the RF vs NoRF groups were due to a lesser influence of the biological drivers and a greater effect of the management drivers of this outcome as days postpartum increased.For reproductive management programs that combine AIE and TAI, the proportion of cows pregnant by a certain time point depends upon the ability of cows to express estrus and the fertility to the first and subsequent AI services.Moreover, first service outcomes are less influential as cows receive more inseminations and have more opportunities to receive TAI.As a result, the influence of factors or events that occur around or immediately after calving on estrous expression and fertility is reduced later in lactation (Stangaferro et al., 2018a;Carvalho et al., 2019).As an example, the magnitude of the effects of clinical health disorders are greater on first than second and greater AI services (Rial et al., 2022).More opportunities to receive TAI services also reduce the detrimental effects of lack of expression of estrus and might increase the fertility of some cows that would otherwise have low fertility if receive AIE. Combining more than one predictor to capture the effects of several biological drivers of expression of estrus and fertility would therefore be expected to help identify cows with larger differences in the percentage of cows pregnant by mid-lactation.This hypothesis was supported as the difference in percentage of cows pregnant by 150 DIM between the E-NoRF and NE-RF group was almost double than that observed for the E-VWP vs NE-VWP or the RF vs No-RF group.Also, in line with expectations, the percentage of cows pregnant by 150 DIM for the groups with a mix of favorable and unfavorable factors (i.e., NE-NoRF and E-RF) was in-termediate suggesting additive effects of AEA and RF.Ultimately, larger differences for the groups generated by combining both factors demonstrated the benefits of this strategy for identifying cows with more extreme performance potential.
Even though data-driven tools and algorithms that enable grouping cows based on levels of multiple predictors at a time are already available and likely to become more ubiquitous in the future, simpler approaches for grouping cows for TRM could be more appealing to dairy farms.Moreover, the herd performance, management, and economic benefits of TRM might be more obvious when fewer groups with larger differences in outcomes of interest are compared.Therefore, we conducted a series of analyses to evaluate the consequences of different cow grouping strategies.We compared differences in performance under conditions in which cows were not grouped, cows were grouped based on a single factor, or the combination of both.As expected, the most favorable (i.e., E-NoRF) and unfavorable (i.e., NE-RF) combinations had the largest and most consistent differences with the whole cohort (i.e., no grouping) for all reproductive outcomes.For some outcomes, the groups created based only on RF or AEA did not differ from the E-NoRF or NE-RF groups; however, in all cases the most contrasting differences with the whole cohort were for the groups created by combining both factors.Collectively, these results suggested that the most effective strategy to identify cows with the largest and most consistent differences in performance compared with the whole cohort of cows (i.e., no grouping) was to combine AEA and RF.A caveat of this approach is that the size of the groups created was smaller than when a single factor was used.A smaller group of cows eligible for targeted management requires larger effect sizes for the interventions implemented to affect whole herd performance or management.
Finally, we compared the performance of the E-NoRF and NE-RF versus the rest of the cows after removing these groups from the whole cohort.This analysis represented a simpler scenario for commercial farms whereby a TRM treatment or intervention is implemented on a single group of cows created based on the combination of more than one factor.Although the differences observed between the E-NoRF and NE-RF groups with the remaining cow cohort were not always greater than those observed when the individual factors were used, the differences were all statistically significant, the effect size was of large magnitude, and importantly, the effects were consistent across all outcomes of interest.Prioritizing consistency across multiple outcomes might come at the expense of potential gains in performance for individual outcomes when a single factor is used.Nevertheless, the proposed strategy combining AEA Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE and RF might be simpler to implement because groups for TRM are created once based on a single approach that generates sizable differences for all outcomes.Thus, managing the E-NoRF or the NE-RF group differently than the rest of the herd might be the most beneficial and yet simple strategy to implement TRM at a commercial herd.
The choice and use of predictors of performance is critical for the design and implementation of TRM strategies (Giordano et al., 2022).Recent research has shown that occurrence of AEA during the VWP was strongly associated with future reproductive potential as cows with and without AEA during the VWP had substantial differences in several reproductive outcomes (Borchardt et al., 2021;Rial et al., 2022).A central hypothesis of this study was that cows grouped based on the combination of data for AEA and other predictors of reproductive performance, would have even larger differences in performance than cows grouped based on AEA or the predictors alone.As there is no established method to select predictors for combining with AEA data, the approach used prioritized inclusion of predictors that were consistently associated with all reproductive outcomes evaluated and could be readily available at commercial dairy farms that use ubiquitous technologies such as automated health and reproductive monitoring systems, milk yield monitoring systems, and genomic testing.Other desired features were automated data collection (e.g., milk yield, rumination time) with no cow manipulation and existing data collection for other purposes to avoid extra costs (e.g., health event records, gDPR).Although the approach and steps followed in this study might be replicated in commercial farms, the approach used to create subgroups of cows might have to be adjusted to individual farms conditions because of the large variability across farms for predictor data available, the number of predictor levels that can be used, and the way in which predictor data can be used.
The first step to ensure inclusion of data with the most predictive value to the RF groups was to evaluate associations between the putative predictors available and outcomes of interest.Overall, our observations from the univariable and multivariable models were generally consistent with previous data and expectations.Health disorders, gDPR, milk yield, and parity were associated with the percentage of cows AIE, P/ AI, and cows pregnant by 150 DIM.Cows diagnosed with at least one health disorder were less likely to receive AIE, had lower P/AI, and fewer were pregnant by 150 DIM.These observations are in line with previous data and the effects of poor health on several biological drivers of expression of estrus and fertility (Pinedo et al., 2020;Manriquez et al., 2021;Rial et al., 2022;Bruinje et al., 2023).We observed better performance across the 3 reproductive outcomes for cows with superior genetic merit for fertility and a hierarchy of performance in line with expectations.This agrees with the greater percentage of cows AIE, greater first service P/AI, and greater percentage of cows pregnant in mid-to-late lactation reported for cows of superior genetic merit for fertility in several recent studies in which cows were managed with programs that combined AIE and TAI (Chebel and Veronese, 2020;Lima et al., 2020;Sitko et al., 2023).Overall, the highest producing cows in this study had the worst reproductive performance.Although associations between milk yield and reproduction are confounded by several factors such as herd milk production level and grouping strategies, our results agree with previous reports of poorer reproductive performance for the highest producing cows within a herd in some studies (Fonseca et al., 1983;Lopez-Gatius et al., 2006;Bello et al., 2012).Our approach to select putative predictors aimed to remove those not as strongly associated with outcomes of interest; however, did not imply lack of value of these predictors for grouping cows for TRM.Indeed, except for the coefficient of variation of prepartum rumination, all other predictors had statistically significant associations with at least one outcome.Thus, these predictors alone in or combination with others could be considered for RF group creation when the goal is to identify cow groups with large differences for individual rather than multiple reproductive outcomes.
Although the association between parity and reproductive outcomes is well-known (Fonseca et al., 1983;Stangaferro et al., 2018a) and it was also the case in this study, we chose not to include parity for creating the RF groups.Our goal was to test a method to group cows that was relevant to primiparous and multiparous cows and the sample size available was insufficient to test hypotheses within parity group.Instead, parity was used to adjust all statistical models.Future studies with larger sample sizes could consider the development of strategies to group cows within parity group.

Study limitations
The most notable limitations of this study were using data from a single commercial farm, a limited sample size for some analyses, and using data from cows enrolled in a randomized controlled experiment comparing different reproductive management strategies.The latter could have generated biases that affected the direction and magnitude of the associations observed.Specifically, the use of different periods of eligibility for AIE after the end of the VWP for cows in the E-VWP and NE-VWP group in the TP-AIE treatment.We also

CONCLUSIONS
We conclude that combining data for predictors collected during early lactation and automated estrus alerts during the VWP can identify subgroups of cows with large differences in reproductive performance potential that could justify implementation of TRM strategies.The value of AEA as a tool for identifying cows for TRM during the VWP might be enhanced when combined with data for cow genetic merit for fertility, health events, and productivity.
Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE if at least one or no AEA were recorded from 21 to 49 DIM, respectively.

Figure 1 .
Figure 1.Data collected for putative predictors of reproductive performance available from 21 d before until the day of calving were rumination time, days in the close-up pen, season of calving, calving-related events (i.e., twins, aborts, stillbirth, and dystocia) and gDPR.Health disorder events (1 to 21 DIM), rumination time (2 to 20 DIM), and milk yield (1 to 49 DIM) data were collected in the postpartum period.Automated estrus alerts (AEA) from an automated estrus monitoring system that used an ear-attached based sensor were recorded from 21 to 49 DIM.Cows were submitted to receive first service with programs designed to prioritize insemination at detected based on AEA and used timed AI for cows not detected in estrus.Cows received AI based on AEA and if not inseminated for up to 31 ± 3 d after the end of a 49 d VWP received timed AI after synchronization of ovulation with an Ovsynch protocol with progesterone supplementation (P4-Ovsynch).gDPR = genomically enhanced daughter pregnancy rate, AEA-VWP = period of collection of automated estrus alerts generated during voluntary waiting period, AEA-AI period = period of insemination based on AEA, TAI = timed artificial insemination, G-P4in = GnRH treatment and intravaginal progesterone releasing device insertion, P-P4out = PGF2α treatment and intravaginal progesterone releasing device removal, P = PGF2α treatment, and G = GnRH treatment.
Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE Table 2. Least squares means (95% CI) for the percentage of cows AI at detected estrus, pregnancies per AI, and percentage of cows pregnant by 150 DIM from multivariable logistic regression models including categorical variables created based on potential predictors of reproductive performance for a cohort study including 886 lactating Holstein cows submitted to first service through a combination of AI at detected estrus and timed percentage of cows inseminated at detected estrus. 2 Diff: value for the referent level minus the value for level in row. 3 P/AI: pregnancy per artificial insemination.4 P150: percentage of cows pregnant by 150 DIM. 5 gDPR: genomic prediction of genetic merit for daughter pregnancy rate.

Figure 2 .
Figure 2. Association between risk factors and estrus during the VWP group with the percentage of cows inseminated at detected estrus (A and B), pregnancies per AI at first service, and (C and D) percentage of cows pregnant by 150 DIM (E and F) for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI and grouped based on records of automated estrus alerts and risk factors for inferior reproductive performance.RF = had at least one risk factor for inferior reproductive performance, NoRF = had no risk factor for inferior reproductive performance, E-VWP = had at least one automated estrus alert during the VWP, and VWP = had no automated estrus alert during the VWP.

Figure 3 .
Figure 3. Percentage of cows inseminated at detected estrus (A) and survival curves for time to first service (B) for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI and grouped based on records of automated estrus alerts and risk factors for inferior reproductive performance.E-NoRF = automated estrus alert and no risk factors for inferior reproductive performance, NE-NoRF = no automated estrus alert and no risk factors for inferior reproductive performance, E-RF = automated estrus alert and risk factors for inferior reproductive performance, NE-RF = no automated estrus alert and risk factors for inferior reproductive performance.

Figure 4 .
Figure 4. Pregnancies per AI at 39 ± 3 d after first AI (A) and percentage of cows pregnant by 150 DIM (B) for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI and grouped based on records of automated estrus alerts and risk factors for inferior reproductive performance.E-NoRF = automated estrus alert and no risk factors for inferior reproductive performance, NE-NoRF = no automated estrus alert and no risk factors for inferior reproductive performance, E-RF = automated estrus alert and risk factors for inferior reproductive performance, NE-RF = no automated estrus alert and risk factors for inferior reproductive performance.

Figure 5 .
Figure 5. Survival curves for time to pregnancy by 150 DIM for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI and grouped based on records of automated estrus alerts and risk factors for inferior reproductive performance.E-NoRF = automated estrus alert and no risk factors for inferior reproductive performance, NE-NoRF = no automated estrus alert and no risk factors for inferior reproductive performance, E-RF = automated estrus alert and risk factors for inferior reproductive performance, NE-RF = no automated estrus alert and risk factors for inferior reproductive performance.

Figure 6 .
Figure 6.Effect of grouping strategies based on favorable (A, B, C) or unfavorable (D, E, and F) factor levels on the percentage of cows AIE, P/AI, and percentage of cows pregnant by 150 DIM for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI.NoGroup = all cows in the cohort, NoRF = no risk factors for inferior reproductive performance, E-VWP = had at least one automated estrus alert during the VWP, and E-NoRF = automated estrus alert and no risk factors for inferior reproductive performance.

Figure 7 .
Figure 7. Percentage of cows that received AIE, pregnancies per AI, and percentage of cows pregnant by 150 DIM for (A) cows grouped based on the most favorable combination of features (E-NoRF) or (B) the most unfavorable combination of features (NE-RF) versus the remaining cohort of cows (RemCohort) for lactating Holstein cows (n = 886) submitted to first service through a combination of AI at detected estrus and timed AI.E-NoRF = automated estrus alert and no risk factors for inferior reproductive performance and NE-RF = no automated estrus alert and risk factors for inferior reproductive performance.
Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE used arbitrary criteria to select and combine potential predictors and then to create the RF and AEA groups.Conducting prospective studies including a larger sample size from several commercial farms with similar reproductive management would improve the external validity of results and could help confirm the direction and strengths of associations observed.Also, controlled trials would be necessary to test the effect on reproductive performance, management, and profitability of TRM interventions applied to cows grouped based on the strategies developed in this study.Establishing structured criteria or methods to identify predictors and groups of cows with different reproductive performance potential based on the combination of multiple factors would favor future research and adoption of TRM strategies.The approaches used in the current study prioritized previous knowledge of biological drivers of reproductive outcomes and use of data that can be easily collected in commercial farms with existing technologies or management practices.

Table 1 .
Rial and Giordano: COMBINING PREDICTORS OF REPRODUCTIVE PERFORMANCE Least squares means (95% CI) for the percentage of cows AI at detected estrus, pregnancies per AI, and percentage of cows pregnant by 150 DIM from univariable logistic regression models including categorical variables created based on potential predictors of reproductive performance for a cohort study including 886 lactating Holstein cows submitted to first service through a combination of AI at detected estrus and timed AI Predictor