Up-to-date evidence on HIV incidence for different populations at various levels of risk is not always easily obtainable. In particular, the populations under consideration would be the people who would benefit from PrEP, i.e. subpopulations at increased probability of HIV acquisition, and not the total U.S. population. Several sources of evidence indicate that utilizing an incidence threshold of 1 % for U.S. subpopulations who would benefit from PrEP is reasonable. A recently published systematic review of global HIV incidence found studies reporting HIV incidence as high as > 2 per 100 PY for groups of PWID and > 4 per 100 PY for groups of MSM in the U.S.; however no data were found for heterosexual women and men who may need PrEP in the U.S. [13].

For MSM, relevant clinical trial data could be from the U.S. component of the HPTN 083 clinical trial [14]. This randomized clinical trial was performed between 2016 and 2020 in multiple settings, including the U.S., and demonstrated superiority of injectable CAB LA, compared with oral daily tenofovir/emtricitabine (TDF/FTC) as PrEP, in preventing HIV. The U.S. HIV incidence for MSM in HPTN 083 was 1.33 per 100 PY in the TDF/FTC arm and 0.26 per 100 PY in the CAB LA arm. These rates are consistent with a baseline incidence without PrEP greater than 1.33 per 100 PY in the U.S. among MSM with indications for PrEP; some studies have estimated an annual HIV incidence as high as 2–7 % among particular subgroups, such as Black MSM or Hispanic/Latino MSM in jurisdictions with high HIV prevalence [13,15–19]. Several large cohort studies and clinical trials of MSM in the U.S. have documented an annual HIV incidence of 1 % or greater for MSM with characteristics that constitute clinical/behavioral indications for PrEP [14,20–22].

There is no recent clinical trial that tested PrEP efficacy among women in the U.S., where the HIV incidence among women is generally much lower than in some international settings. A recently published pooled analysis of PrEP efficacy from 11 post-approval studies of TDF/FTC conducted across many settings provides an estimated HIV incidence for women with no or consistently low adherence to PrEP, which can approximate incidence among women with indications for PrEP but not receiving PrEP. This incidence was 1.27 per 100 PY, compared with 0–0.13 per 100 PY among women with consistently high or consistently daily adherence to PrEP (90 % effectiveness). There was very little U.S. representation among the studies included in this pooled analysis, however [23]. Groups of women with estimated annual incidence ranging from 0.32 % to 1 % per year have been described among a few U.S. cohorts with clinical or behavioral characteristics that confer indications for PrEP, however one study was published over 20 years ago [21,24]. Race and ethnicity, as well as geography, are other factors associated with elevated HIV annual incidence [30,36]. As of November 2024, among 2556 U.S. women who used an HIV self-test obtained through CDC’s ongoing “Together Take me Home” project and who responded to a follow-up survey, 1.3 % reported a new positive HIV self-test result (personal communication, CAPT. R. MacGowan).

The Bangkok Tenofovir trial was, to our knowledge, the only trial of PrEP efficacy in preventing HIV among PWID; it was conducted between 2005 and 2010 in Thailand [25]. The HIV incidence without PrEP among PWID in this study was 0.68 per 100 PY vs. 0.35 per 100 PY in the PrEP arm (49 % efficacy); for those with detectable levels of tenofovir efficacy was 74 %. Estimates by Evans et al [26] indicate high incidence of HIV (9.3 per 1000 PY) among groups of PWID in U.S. settings.

A method to estimate the “need for PrEP” in the U.S. could be based on the concept of the “number needed to treat”, a metric estimating the number of people who need to receive a health intervention to gain benefit over a period of time e.g., prevent an illness or death, in this case PrEP for preventing new HIV infections [4]. The number needed to treat (NNT) is typically derived from clinical trials or other rigorously designed studies that evaluated a particular intervention in a given setting. When a study outcome is expressed as a percent risk reduction, the number needed to treat is the inverse of the absolute risk reduction. Similarly, when a study outcome is based on time of exposure (patient-years, PY), the number needed to treat is calculated based on cumulative event proportions and reflects additional person-years needed to treat. NNT is a relatively simple metric and has the advantages of being easy to understand and applicable to public health efforts. However, it is dependent on the population studied, and will be lower for groups at higher baseline risk and higher for groups at lower baseline risk [12]. Additionally, different PrEP modalities will have different NNT, based on the effectiveness of the intervention (which depends on adherence); injectable long-acting cabotegravir (CAB LA) and lenacapavir will likely be associated with a somewhat lower NNT, compared with oral daily or event-driven PrEP, given the same HIV incidence, due to their higher effectiveness [13].

The NNT can be defined as follows:

NNT=1IRU-IRT=1IRU×EfficacyPrEP, rounded up to the nearest whole number, where IRT is the HIV Incidence Rate in persons who received PrEP and IRU is the HIV Incidence Rate in persons who did not receive PrEP (or, in this case, the baseline HIV Incidence in those who would benefit from PrEP).

Based on a baseline HIV incidence of 1 % per year and assumed PrEP effectiveness in preventing HIV acquisition due to sexual exposures of 90 % with high adherence, as shown in multiple reports [27–32], the NNT with PrEP to prevent one additional HIV infection in that population would be 1IRU×EfficacyPrEP=10.01×0.9=112. As mentioned, NNT varies by level of incidence within a population, and it will vary by particular transmission groups, sexual networks, age, and other factors. [12,13,16]. NNT will also vary by PrEP modality, however, given that the great majority of current PrEP use in the U.S. is of oral PrEP, and CAB LA use is currently less than 2 % [33], we based the NNT calculations on oral PrEP effectiveness.

For PWID, assuming an incidence of 1 % per year as a threshold for the PWID subpopulation who may need PrEP and with an assumed PrEP effectiveness for PWID of 70 %, the NNT with PrEP to prevent one additional HIV infection would be 143.

Population-level “need for PrEP” (PPN) can be estimated by applying the number needed to treat (NNT) to the most recent and locally granular available data of new HIV infections obtained by HIV surveillance [34]. Therefore, PPN would be calculated as: PPN=NNT×N, where N denotes the number of new HIV infections in a year, observed within a population whose baseline HIV Incidence Rate corresponds to the IRU used to calculate the NNT.

Tthis calculation assumes that all new HIV infections occurred among people in subpopulations with annual incidence of HIV of 1 % or greater, an assumption that may not be true. For any population, there is a distribution of people at risk for HIV, some with higher underlying risk (greater than 1 %) and some lower. If only a proportion of people with new HIV infections are in subpopulations with a baseline annual incidence of 1 % or higher, then we need to remove the proportion of new HIV infections occurring within the subpopulations corresponding to a baseline annual incidence below 1 % from the PPN calculation. Therefore, the formula for PPN is modified as follows:

If p denotes the proportion of new HIV infections that occur among persons within subpopulations with annual incidence of 1 % or higher, then the PPN would be calculated as: PPN=NNT×N×p.

Several cohort studies in the US have demonstrated that people with epidemiologic, clinical or behavioral factors that confer a higher probability for HIV acquisition have annual HIV incidence of around 1 % or higher [20–22,24,35–39]. Some of these studies, as well as findings from CDC’s National HIV Behavioral Surveillance (NHBS) have also examined the prevalence of such factors among people with new HIV diagnoses. For example, condomless anal sex with 2 + partners has been found in approximately 70–80 % of MSM with a new HIV diagnosis, whereas rectal gonorrhea or chlamydia (GC/CT) infection has been found in a wider range of proportions (26–68 %) in different cohorts of MSM with a new HIV diagnosis [20–22,38,40]. Condomless anal sex (CAS) increases the rate of HIV acquisition 2–3 fold [21,22,38]; given an overall annual incidence rate among MSM of about 0.58 %, CAS may then correspond to an annual incidence of ≥ 1 %. The prevalence of CAS with two or more partners among MSM with a new HIV diagnosis from publishedstudies mentioned above as well as from NHBS data (Dr C. Wejnert, personal communication) can aid in determining that the proportion of new infections acquired by MSM at ≥ 1 % annual probability of HIV acquisition could be about 0.7; while about 30 % of new HIV infections stem from MSM subpopulations with a lower than 1 % per year risk of acquiring HIV.

For heterosexual women in the US, the factors most predictive of higher HIV incidence in different cohorts included substance use and demographic characteristics such as race and geographic location; injecting drugs and crack cocaine use was associated with elevated HIV incidence rates up to 30-fold [35], whereas sexual behavioral factors alone did not confer annual HIV incidence rates of 1 % or higher [21,24,35,39,41]. The prevalence of substance use varied between studies of women with recent HIV infection from 14 % to 60 %; data were based on small numbers of women with new HIV diagnosis [24,35,39]. A reasonable approximation of the proportion of new HIV infections that came from a subpopulation of women with a 1 % or higher HIV incidence rate may be 0.3, near the midpoint of the estimates of the studies mentioned.

As there were no relevant data we could find for heterosexual men in the U.S., we selected 0.2 as the proportion of new HIV infections that come from a subpopulation of heterosexual men with an underlying HIV incidence rate of 1 % or higher. These proportions set for both heterosexual women and men were somewhat arbitrary selections and could likely vary.

Sharing syringes and condomless sex in the past 12 months were associated with elevated proportions of recent HIV infections by 1.7–2.8 fold, respectively, in a study among PWID [37], where the overall incidence of recent HIV infections was 0.6 %. Among the 50 people with new HIV infections, 42 % shared syringes and 70 % reported condomless sex in the past 12 months. Therefore, we opted to use 0.5 as the proportion of new infections coming from a subpopulation at ≥ 1 % riskfor PWID.

A large proportion of new HIV infections (approximately 70 %) among MSM are attributed to a subpopulation with ≥ 1 % annual HIV incidence, so calculations using the NNT of 112 corresponding to an underlying incidence of exactly 1 % will overestimate the NNT because the incidence among this group is likely higher than 1 %. To improve upon this calculation, we can use available epidemiologic information to identify different subgroups of MSM corresponding to different levels of underlying annual incidence ≥ 1 %.

For PPN calculations, the proportion p can be divided into n subgroups, each with a different incidence rate and NNT:

p1,p2,p3,…,pn=proportionofnewHIVinfectionsineachsubgroup, where ∑i=1npi=p

IR1,IR2,IR3,…,IRn=HIVincidencerateineachsubgroup

NNT1,NNT2,NNT3,…,NNTn=NNT for each subgroup, where NNTi=1IRi×EfficacyPrEP PPN=∑i=1nNNTi×N×pi

Rectal GC/CT is associated with an increase in the rates of HIV acquisition 3–4 fold, which may place the annual HIV incidence at 2 % or higher for those with rectal GC/CT [20,40,42,43]. Several studies indicate that rectal GC/CT are associated with approximately 15 % (and up to 30 %) of all incident HIV infections in the U.S. [40,44]. Based on this, we assumed that approximately 20 % of incident HIV infections correspond to subpopulations of MSM with annual HIV incidence of ≥ 2 %.

Given this epidemiologic evidence, for MSM we can create two subgroups: Subgroup 1 with high incidence (1 %) and subgroup 2 with very high incidence (2 %, estimated as those with rectal STI, representing 20 % of all infections). Based on these two groups, the adjusted NNT for subgroup 1 will remain 112 corresponding to an underlying incidence rate of 1 %, while the NNT for subgroup 2 will be 56 corresponding to an underlying incidence rate of 2 %, given the same PrEP effectiveness of 90 % in both subgroups. These estimated NNTs are supported by modeling studies. Compartmental models of the HIV epidemic at both the national level and focusing on a Southern state estimate the NNT for MSM to be roughly 100 and 50, respectively (C. Lyles, personal communication). Another modeling study of the U.S. HIV epidemic among MSM that included social network dynamics in various MSM populations and incorporated risk behavior, sexual partnerships, and other HIV transmission factors, also supports these estimated NNTs [16]. It should be noted that HIV incidence varies by race/ethnicity as well as age, with Black and younger MSM having higher incidence [34]; thus such groups of MSM may differ in the distribution of proportions at “high” vs “very high” incidence, and the corresponding incident cutoffs may differ as well.

For heterosexual men and women and for PWID there is no additional evidence we could find to help differentiate subgroups at higher levels of HIV incidence. Thus, the PPN is calculated based on an underlying incidence of 1 % and the corresponding NNT and p for each transmission groups, as discussed earlier.

Based on above formulas and assumed parameter values, and utilizing the estimated number of new HIV infections in 2022 from the latest HIV surveillance report [34], the estimates of the magnitude of the Population Need for for PrEP (NNP) in 2023 in the U.S. are (Table 1):

PPN among MSM: NNT1,MSM×NMSM×p1,MSM+NNT2,MSM×NMSM×p2,MSM=(112×22,500×0.5)+(56×22,500×0.2)=1512,000

By adding these PPN estimates across the different transmission groups, the approximate total number of additional people who need to receive PrEP in the U.S. to prevent the new HIV infections that occurred in 2022 in the subpopulations with underlying annual incidence of 1 % or higher would be estimated as 1888,130.

Table 1 provides a range of PPN estimates for a range of PrEP effectiveness (50 %−90 %) [13–18]).

As the number needed to treat calculates the number of additional people needed to be treated in a year to prevent one infection, we need to also take into account the number of people who were already using PrEP for our total PPN estimate. Given that 366,359 people received PrEP prescription in the U.S. in 2021 [4], and assuming they corresponded to an underlying HIV incidence of ≥ 1 % per year, this number can be added to the PPN estimate above, so that the total PPN estimate becomes PPN= 2254,489.

These calculations can also be made for discrete jurisdictions for which there are available HIV incidence data (states, counties) and for different sex, age, and race/ethnicity groups. Jurisdictions can then estimate the number needed to receive PrEP to achieve their prevention goals. This can be very useful from a programmatic perspective, and for monitoring progress towards achieving public health goals. Table 2 provides 2023 PrEP estimates based on this method for each U.S. state, using the same assumptions as described above. The method can also be applied to distinct demographic groups for which there are data on new HIV infections (or, for jurisdictions for which this number is not available, the number of diagnoses could be used as a proxy).

The method we have developed for estimating the Population Need for PrEP (PPN) can be used with different levels of HIV incidence as thresholds; thresholds could also vary for different transmission groups or localities. Based on an annual HIV incidence rate of 1 % as our defined threshold for PrEP ‘need’, our method estimated the number who need to be prioritized for PrEP in the U.S. as 2254,489 people in 2023. It is thus higher than the estimate of the currently used PrEP coverage indicator (which estimated approximately 1.2 M people with indications for PrEP in the U.S. in 2015). Advantages of the PPN method are its relative simplicity, and the fact that it can provide PrEP need estimates per jurisdiction and per any demographic or transmission group for which HIV incidence data are available; such data are broadly available and frequently updated in National HIV Surveillance. Through the possible inclusion of multiple subgroups within any group, the method can also accommodate information on these subgroups at different levels of HIV incidence which account for differing proportions of new HIV infections (population attributable fractions)when data are available. Local policy makers can use the estimates we used in this paper to define the size of subgroups at different levels of HIV incidence, or they can utilize local data that may be tailored to specific demographic or transmission groups. The PPN estimate can function as the denominator for a PrEP indicator and its findings would be valid as long as they represent a reasonably accurate reflection of the current state of the epidemic in the particular locality. Our estimates focus on preventing the portion of new infections that occur among those subpopulations with higher underlying annual HIV incidence (≥ 1 %). To be successful in preventing these infections, programs should prioritize PrEP for these communities or subpopulations with a 1 % or higher risk of HIV acquisition. Limitations of the method include: i) It is difficult to accurately estimate HIV incidence in the different transmission groups in the U.S., as well as the proportions of new HIV infections that correspond to subpopulations with differing levels of incidence ≥ 1 %. ii) We applied measures of PrEP effectiveness based on relatively high adherence. Effectiveness in practice may be lower, which would increase the PPN. Given that PPN represents the number who need to be on PrEP for a given time period, in this instance a year, this limitation could be overcome if use of PrEP (the numerator of a PrEP indicator) incorporates a measure of PrEP consistency and persistence, such as time covered by a PrEP prescription. iii) This method of estimating PrEP “need” does not provide information about a particular individual’s need for PrEP; clinical evaluation is necessary to guide individual recommendations; iv. We could be overestimating PrEP “need” if the proportion of new infections that occur among people with annual incidence of ≥ 1 % is lower than what we estimated based on clinical and epidemiologic evidence. This would mostly apply to heterosexual or PWID groups, for which incidence data are limited. We could also be overestimating among subpopulations of MSM, if the HIV incidence rates among those subgroups were higher than what we estimated above based on the 1 % and 2 % underlying HIV incidence rate thresholds we assumed. Finally, HIV incidence is estimated in National Surveillance and as such it has a range of uncertainty.

In summary, we have provided initial estimates of U.S. PrEP “need” based on an alternative method using NNTs for different transmission groups. This method could be further modified using different thresholds for HIV incidence to define PrEP “need”. It could also be further refined for specific population groups and jurisdictions, based on additional information of PrEP use patterns as well as parameters of the HIV epidemic locally.