AGING AND MORTALITY OF PERSONS WITH HIV: A NOVEL

K ALMAN F ILTERING AND DMD FRAMEWORK

Alex Viguerie Elisa Iacomini

Department of Pure and Applied Sciences Department of Environmental and Prevention Sciences
arXiv:2503.16297v1 [math.DS] 20 Mar 2025

Università degli studi di Urbino Carlo Bo Università degli studi di Ferrara

Urbino, PU (Italy) 61029 Ferrara, FE (Italy) 44121
[email protected] [email protected]

A BSTRACT
Due to the widespread availability of effective antiretroviral therapy (ART) regimens, average
lifespans of persons with HIV (PWH) in the United States have increased significantly in recent
decades. In turn, the demographic profile of PWH has shifted. Older persons comprise an ever-
increasing percentage of PWH, with this percentage expected to further increase in the coming years.
This has profound implications for HIV treatment and care, as significant resources are required
not only to manage HIV itself, but associated age-related comorbidities and health conditions that
occur in aging PWH. Effective management of these challenges in the coming years requires accurate
modeling of the PWH age structure. In the present work, we introduce several novel mathematical
approaches related to this problem. We present a workflow combining a PDE model for the PWH
population age structure, into which publicly-available HIV surveillance data is assimilated using the
Ensemble Kalman Inversion (EKI) algorithm. This procedure allows us to rigorously reconstruct the
age-dependent mortality trends for PWH over the last several decades. To project future trends, we
introduce and analyze a novel variant of the Dynamic Mode Decomposition (DMD), nonnegative
DMD. We show that nonnegative DMD provides physically-consistent projections of mortality and
HIV diagnosis while remaining purely data-driven, and not requiring additional assumptions. We then
combine these elements to provide forecasts for future trends in PWDH mortality and demographic
evolution in the coming years.

Keywords HIV · population-structured models · inverse ensemble Kalman filter · Dynamic Mode Decomposition

1 Introduction

In recent decades, the life expectancy of individuals living with HIV, particularly in industrialized countries, has
increased significantly, largely due to advancements in antiretroviral therapy (ART) [1, 2]. ART has transformed HIV
from a fatal disease into a manageable chronic condition, allowing many persons with HIV (PWH) to live lifespans
nearly commensurate with the general population [1, 3, 4]. As a result, the demographic profile of persons with
diagnosed HIV (PWDH) in the United States has shifted. As shown in Fig. 1 persons aged 55 and older have gone
from approximately 16% of the PWDH population in 2008 to nearly 45% in 2022 [5]. Such a drastic demographic shift
presents new challenges in the fight against HIV, as long-term management of chronic HIV infection will require an
increasing share of resources.
The growing population of older PWDH requires the healthcare system to address not only the ongoing management
of HIV itself, but, increasingly, comorbidities and age-related conditions that arise as this population ages. These
include cardiovascular disease, diabetes, osteoporosis, and certain cancers, which occur at higher levels, and at younger
ages, among persons with HIV as compared to the general population [6, 7, 8, 9, 10, 11, 12, 13]. Older persons
with HIV also may suffer from cognitive and functional impairment [14, 15]. Additionally, the high burden of health
comorbidities leads to high levels of polypharmacy among older PWDH [16]. Finally, although ART has significantly
increased overall lifespan among PWDH, ART itself is associated with health complications arising from long-term
 PRIME AI paper

Figure 1: Age distribution of the diagnosed PWH population in the United States over the years 2008-22.

use [17, 16, 18]. As a result, healthcare providers must adapt their approaches and strategies to encompass a broader
spectrum of health issues among persons with HIV [11, 19].
Given the changing landscape of HIV care, the focus of public health strategies must also evolve. Prevention efforts
now must include the mitigation of health complications associated with HIV in older adults, long-term HIV infection,
and long-term ART use among PWDH [11, 19], necessitating a comprehensive approach that integrates HIV care with
general healthcare for chronic disease prevention and management [13, 19, 11]. Effective strategies will be crucial to
maintain the health and quality of life of the aging PWDH, ensuring that the progress made in extending life expectancy
is not undermined by the rise of preventable chronic conditions, and that increased life-years are not accompanied by
decreased life quality. Furthermore, despite decreases in new HIV infections and diagnoses in recent years, improved
life expectancy among PWDH has led to a significant increase in the PDWH population as a whole in the United States
and other high-income countries [20, 21], compounding aging-related difficulties, as HIV care must be allocated across
a large, and increasing, population.
To help address these challenges, the development of accurate mathematical models capable of providing reliable
projections for the PWDH age structure in coming years is an important priority in HIV policy research. Such models
can play a fundamental role in the planning and evaluation of intervention strategies and the allocation of valuable HIV
treatment and prevention services. Recent work on this topic includes [22], which applied a compartment model for
demographic projections, and [23, 24], which used an agent-based model to project how demographic changes in the
US PWDH population may affect the burden of comorbidities. A particularly important component of these models is
their ability to accurately characterize age-dependent PWDH mortality rates, and furthermore, how they may change in
coming years. This arguably forms the foundation of such models, as the overall cost of HIV-related care depends most
heavily on total PWDH person-years.
Given the importance of the topic, in the present work, we seek to estimate how the age-dependent mortality rates for
PWDH in the US have changed in recent years in a mathematically rigorous manner, and furthermore, extrapolate from
these data to project future such changes. To the authors’ knowledge, no such study has been undertaken, certainly
not to describe changes in mortality at the age- and time-continuous level. To accomplish this task, we employ a data
assimilation approach, in which age-discrete HIV surveillance data is integrated into a partial differential equation
(PDE) model via an inverse ensemble Kalman filter (EnKF) to reconstruct the time evolution of age-dependent PWDH
mortality since 2009. To project future trends in age-dependent mortality based on our reconstructions, we develop and
apply a novel variant of Dynamic Mode Decomposition (DMD), nonnegative DMD. By incorporating these data into
our PDE model, we provide a forecast PWDH age structure, annual HIV diagnoses, and annual PWDH deaths in the
United States over the coming years.
To reconstruct the evolution of age-dependent PWDH mortality rates in recent years, we employ the Ensemble Kalman
Inversion (EKI). The EKI is an adaptation of the ensemble Kalman filter (EnKF), a powerful tool for estimating the
state of a dynamical system by iteratively refining predictions based on noisy observations. Originally introduced in the
1990s [25], the EnKF has gained widespread use due to its ability to fuse model predictions with measurement data.
Ensemble Kalman filtering has been employed to enhance the accuracy of solutions in diverse applications, including

2
 PRIME AI paper

oceanography [26], reservoir modeling [27], weather forecasting [28], milling process [29], geophysical applications
[30], physics [31] and machine learning [32, 33]. In public health, the EnKF was recently used to forecast heroin
overdose deaths [34]. The EKI is a modified version of the EnKF designed for inverse problems, wherein underlying
model parameters are estimated given a time-series of observations corresponding to model states [35, 36, 37]. The
EKI is attractive computationally, as it does not require the computation of derivatives of the underlying model, as well
as mathematically, as it allows for provable convergence results [35, 36].
To project future trends from the reconstructed mortality data, we introduce a novel modification of the Dynamic
Mode Decomposition (DMD), nonnegative DMD. DMD is a Scientific Machine Learning (SciML) technique that
extracts the most relevant dynamical structures existent in time-series data using a purely data-driven approach, with
applications ranging from short-time future estimates to control, modal analysis and dimension reduction [38]. DMD
has been deployed in a wide range of scientific and engineering applications such as biomechanics [39], oncology [39]
epidemiology [40, 41, 42], climate modeling [43], aeroelasticity [44], additive manufacturing [42], urban mobility [45],
and the modeling and simulation of batteries [46]. The authors are not aware of any work using DMD to forecast the
evolution of mortality rates, or similar quantities. In general, however, standard DMD fail to respect non-negativity
when used for forecasting beyond the training period, even when the underlying training data are nonnegative [47, 42].
This makes DMD in its original formulation potentially unsuitable for forecasting nonnegative quantities, such as
mortality rates, and as such must be modified accordingly.
The article is outlined as follows. In the methods section, we provide a high-level description of the mathematical model
and the EnKF method. We then introduce the proposed novel variant of DMD, nonnegative DMD, discuss its numerical
solution, and establish some of its basic mathematical properties. We then present the results of our analyses. We first
present the reconstructions of age-dependent PWDH mortality curves over previous years. We then use nonnegative
DMD to obtain projections of future mortality rates, and apply these rates in our underlying PDE model in order to
provide projections for age-dependent PWDH prevalence and mortality through 2030. We then provide a detailed
discussion of the implications of our findings and conclude by discussing possible future directions to extend this
research.

2 Methods
2.1 Mathematical model

We consider an age-structured PDE model for the population u(a, t), which gives the size of the PWDH population u
aged a at a time t. The basic equations read:
∂u ∂u
+ = −µ(a, t)u + λ(a, t) for t0 < t ≤ tend , 0 ≤ a < ∞,
∂t ∂a (1)
u(a, 0) = u0 (a) at t = t0 , u(0, t) = g(t),
where µ(a, t) ≥ 0 is the age- and time-dependent mortality rate, u0 ≥ 0 is the age-distribution of the PWDH population
at the initial time, and λ(a, t) ≥ 0 gives the number of new HIV diagnoses of persons aged a at a time t, and is known
from data.
The inflow condition g(t) ≥ 0 defines new births, and may be time- and/or state-dependent in general. In the present,
however, we assume g(t) = 0, for several reasons. First, levels of perinatal HIV in the US are very low [48]. Second,
publicly available HIV surveillance data in the US does not include persons younger than 13 years of age [20]. Finally,
the method of generating continuous age-distributions from discrete age-bins available from surveillance data (discussed
in section 2.2), used to generate u0 (a) and λ(a, t), ensures u(a, t) > 0 and remains smooth for all a > 0 throughout
the simulation period.
In practice, we solve (1) numerically, considering a bounded age domain 0 ≤ a ≤ 101. We employ a second-order
backward differentiation (BDF2) scheme throughout the age domain, assuming physically consistent ‘ghost node’
values of at−1 = at−2 = 0 for all t. For the time discretization, we use Heun’s method to solve the first time step, and
BDF2 for the remaining time steps.
Models of this type are common in many applications in the natural and social sciences, and we refer the reader to [49]
for a presentation of these models and [50] for an extensive discussion of more complex extensions.
Denote the domain of u(·, t), the set of admissible human ages, as A = R+ = [0, ∞), endowed with a σ-algebra B
and nonnegative measure ν(A). In the following, we use u(a, t) to denote a generic solution of (1) and provide the
following positivity result:
Theorem 1: If u0 ≥ 0 for all a and µ ≥ 0, λ ≥ 0 for all (a, t), then u(a, t) ≥ 0 for all (a, t).
Proof. Since the age a advances in sync with time, we can consider the characteristic curve a(t) = a0 + t for some

3
 PRIME AI paper

fixed a0 , and note that da/dt = 1. Along the characteristic, we can rewrite (1) in terms of a function v(t), depending
only on t:
v(t) := u(a(t), t) = u(a0 + t, t). (2)
Differentiating with respect to t gives:
dv ∂u ∂u d ∂u ∂u
= + [a0 + t] = (a0 + t, t) + (a0 + t, t). (3)
dt ∂t ∂a dt ∂t ∂a
where the last equation on the right hand side follows from da/dt = 1. From (2), (3) along the characteristic (1) reduces
to the ordinary differential equation:
dv
= −µ(a0 + t, t)v(t) + λ(a0 + t, t). (4)
dt
Rt
µ(a0 +τ,τ )dτ
Re-arranging, multiplying by the integrating factor e 0 , and integrating from 0 to t:
Rt
Z t Rξ
µ(a0 +τ,τ )dτ µ(a0 +τ,τ )dτ
v(t)e 0 = v(0) + λ(a0 + ξ, ξ)e 0 dξ. (5)
0

Recalling that v(0) = u0 (a0 ):

Rt
 Z t Rξ

u(a0 + t, t) = e− 0
µ(a0 +τ,τ )dτ
u0 (a0 ) + λ(a0 + ξ, ξ)e 0
µ(a0 +τ,τ )dτ
dξ . (6)
0

From our assumptions on µ, u0 , λ, all terms on the right-hand side are nonnegative, and hence:
u(a0 + t, t) ≥ 0 ∀t. (7)
Furthermore, since a0 was arbitrary, (7) holds for any a0 ≥ 0, implying:
u(a, t) ≥ 0 ∀(a, t), (8)
as was to be shown.

2.2 Ensemble Kalman Inversion

The Ensemble Kalman Inversion (EKI) is a computational technique that extends the principles of the Kalman filter
to solve inverse problems by utilizing an ensemble of model realizations. By iteratively updating this ensemble with
observational data, EKI effectively estimates unknown parameters and reduces uncertainty, making it particularly
suitable for high-dimensional and nonlinear inverse problems [35, 36, 37].
Here, we describe the Ensemble Kalman Inversion procedure applied in the current work. In order to do this, let us
introduce the notation we will use in the following:

• The overline symbol · denotes an averaged quantity

• u(a, t) denotes the age-distribution of the PWDH population in time;
• µ(a, t) is our unknown; in this case, it defines the time- and age-dependent mortality curve;
• Γ is the matrix of the covariance of the noise, which is assumed to be known;
• G(·) is our output of the total annual PWDH deaths in each age bracket, calculated from the population state u
and mortality curve µ.
• nkf is the number of Kalman filtering steps.
• A(µ, λ) is the operator which advances population age distribution ∆t units in time for a given µ(a, t) and
λ(a, t) according to the system dynamics defined by (1):
u(a, t + ∆t) = A (µ(a, t), λ(a, t)) u(a, t). (9)

For the sake of readibility, we hereafter write µ(a, t), u(a, t), and λ(a, t) respectively as µt , ut , and λt , with the
dependence of each quantity on age a understood if not explicitly denoted.
We seek to use the EKI to estimate the time- and age-dependent mortality curve µt , which is not observable directly,
by incorporating information from the model (1) and surveillance data. The estimation of µt relies on the current age
distribution Pt , calculated from the (1), and the age-bracketed annual mortality data yt , available from surveillance
data [5]. To ensure regularity in our estimate, we do not estimate µt at each age individually, but compute its value at

4
 PRIME AI paper

ages 1, 10, 22, 32, 42, 52, 57, 62, 67, 72, 77, 82, 95, 101. We then consider the full curve µt as the piecewise cubic
Hermite interpolating polynomial through these points. We note that we performed a sensitivity analysis to the choice
of interpolation points, and found that the effect on the overall reconstruction was not large, provided that there are
sufficiently many points across all age ranges.
We begin by initializing an ensemble of mortality curves at µ1t,j for j = 1, ..., J at t = 2009, by sampling from the
distribution of the mortality rate of the general (non-PWH) population µ0 . The superscript number refers to the current
filter step, further clarified in the following. We assume a multivariate normal distribution with mean µ0 and covariance
diag(µ20 ), the component-wise square.
Similarly, we also initialize an ensemble of population age structures Pt0 ,j for j = 1, 2, ..., J at the initial time
t0 = 2009, which serve as the initial conditions for (1). Each Pt0 ,j is generated from surveillance data for the diagnosed
PWDH population at year-end 2008 [5]. These data are stratified into age brackets: 13-24, 25-34, 35-44, 45-54, 55-59,
1
60-64, 65-69, 70-74, 75-79, 80-84, and 85+. We define each Pt,j such that the overall number of PWDH in each age
bracket is equal to the surveillance data, while the distribution of PWDH within each age bracket are defined with a
uniform sample. For each calendar year ti = 2009 − 22, i = 1, 2, ... during the simulation, we follow an identical
procedure using the age-bracketed annual new HIV diagnosis data from [5] to define an ensemble λti ,j , j = 1, 2, ..., J
for new entries in the model (1).
Assume that (1) is discretized time with time step ∆t = 1/m, m ∈ N, such that each year is divided into m time-steps.
For each simulated calendar year ti , we perform nkf inverse Kalman Filtering steps. The procedure is given in
Algorithm (1). In a given year, we perform nkf filtering steps. At the nn-th filtering step, we first advance the Pjnn
following (1) by solving the model over m time-steps, using λti ,j and the mortality curve µnn ti ,j . Note that we must
keep a running update of the simulated age-bracketed mortality G(Pjnn , µnn
ti ,j ) at each time step, so that an end-of-year
total is available.
If nn is less than the total number of filter steps nkf , we use the calculated G(Pjnn , µnn
ti ,j ) and the surveillance mortality
data yti to obtain the updated mortality curve µnn+1 ti ,j . We then simulate the model for the year ti again, using with
nn+1 1 nkf
the updated µti ,j . If nn = nkf , we advance to year ti+1 , setting µti+1 ,j = µti ,j and generating λti+1 ,j from the
surveillance data for year ti+1 . Note that the solution of the EnKF procedure is given by the mean of the ensemble at
the step nkf for each year.

Remark. Note that, as (1) must be solved nkf times in each year, we must keep values of Pj from the pre-
ceding year in order to repeatedly solve the first time-step of the current year. The exact number of time-step(s) to be
saved depends on the specific time-stepping scheme used. As we used BDF2 in the current work, we kept the last two
time-steps from the nkf -th filtering step of the previous year.

2.3 Nonnegative DMD

In the present work, we introduce a novel variant of DMD, nonnegative DMD, and use it to extrapolate the evolution
of the age-dependent mortality rates (calculated using the Ensemble Kalman Inversion) and the age structure of new
HIV diagnoses (estimated from surveillance data) through the year 2030. In the following, we will provide a brief
introduction to the standard DMD algorithm, then we will introduce the nonnegative DMD algorithm, and provide some
information regarding its calculation.

2.3.1 Standard DMD

In this subsection, we provide a brief description of DMD. For the purposes of simplicity and exposition, we restrict
our discussion to the original formulation of DMD as proposed in [51]. We remark that this is a vast topic of research,
and refer the reader to [38] for a more comprehensive treatment of DMD. Suppose we have time series consisting of n
snapshots:
n
xi ∈ Rm ∀i.

xi i=1 , (10)
We arrange these snapshots into two m × n − 1 matrices X1 , X2 column-wise as follows:
| | | | | |
" # " #
X1 = x1 x2 ... xn−1 , X2 = x2 x3 ... xn . (11)
| | | | | |
DMD seeks to reconstruct the operator A mapping X1 to X2 , that is:
AX1 = X2 , (12)

5
 PRIME AI paper

Algorithm 1 Algorithm with fixed number of filter steps for the model G
1: Initialize state ensemble ut0 ,j
2: Initialize measurement noise: Q = diag(Var(G(ut0 , µ0 )), a diagonal matrix consisting of the variance in observed
deaths in each age group for the general-population mortality µ0 over the initial state ensemble ut0 .
n 2 2
3: Initialize the mortality curve ensemble µt0kf
,j ∼ N (µ0 , diag(µ0 )), with j = 1, 2, ... . . . , J, where µ0 is interpreted
as the element-wise square, and nkf > 0 the number of optimization steps
4: Set n = 0, t1 = 2009
5: for ti =2009:2022 do
6: Generate ensemble of annual new diagnoses λti ,j ;
nkf
7: Set µ1ti ,j = µti−1 ,j ;
8: for nn=1:nkf do
9: Set G(uj , µnn
ti ,j ) = 0;
10: for k = 1 : m do
11: Advance the population structure ensemble and age-structured mortality output:
uti +k∆t, j = A(µnn
ti ,j , λti ,j )uti +(k−1)∆t, j ;
G(uj , µnn nn nn
ti ,j ) = G(uj , µti ,j ) + ∆t diag(µti ,j ) uti +k∆t, j ;

12: end for P

J
13: µ̄nn = J1 j=1 µnnti ,j
nn J
G = J1 j=1 G(uj , µnn
P
14: ti ,j )
15: η ∼ N (0, Q)
16: Solve the EnKF procedure:
J
1 X nn nn
C(µnn ) = (µ − µnn ) ⊗ (G(uj , µnn
j )−G )
J j=1 j
J
1X nn nn
D(µnn ) = (G(uj , µnn
j )−G ) ⊗ (G(uj , µnn
j )−G ) (EnKF)
J j=1
−1 −1
µnn+1 = µnn nn nn
yti + η − G(uj , µnn



j j + C(µ ) D(µ ) + Γ j )

17: end for

18: end for

which can be obtained as [51]:

†
A = X2 (X1 ) , (13)
where X1†
is the Moore-Penrose pseudoinverse of X1 (see e.g. [52]). This formulation of A solves the minimization
problem:
arg min ∥AX1 − X2 ∥F , (14)
A

with ∥ · ∥F denoting the Frobenius norm [53]. Note that, in practice, A can be large and dense, and is not typically
computed directly as (13). Instead, algorithms which efficiently compute lower-rank approximations of A can be
leveraged to avoid ever forming A explicitly (see e.g. [38, 51, 39]).

2.3.2 Nonnegative DMD algorithm

Being a completely data-driven algorithm, DMD, particularly when extrapolated past the training interval, will not
necessarily respect the physics of the underlying systems. For example, in [46], it was demonstrated that standard
DMD can fail to predict periodic limit cycles. Similarly, in [42], it was shown that DMD, when formulated as (13) (or
with a lower-rank approximation), will conserve the mass of a mass-conservative system (assuming the data is also
mass-conservative), even projected out in time; however, other key physical properties, in particular nonnegativity, were
not preserved in the extrapolations.
In the present work, we employ DMD to project two strictly nonnegative physical quantities: future HIV diagnoses and
future age-dependent mortality rates among PWH. While standard DMD reliably reproduces the training data in both

6
 PRIME AI paper

instances, attempting to extrapolate beyond the training interval results in negative, oscillatory reconstructions, even
over short projection windows (1-2 years).
In order to guarantee (An x)i ≥ 0, ∀ i, n, assuming a nonnegative x, it is enough to guarantee that the matrix A itself
is nonnegative. The most obvious solution would be to find A as in (13), and then use nonnegative matrix factorization
to replace A with an approximate nonnegative version A ≈ A b = M N , that is:

arg min ∥M N − X2 (X1 )† ∥F , Mi,j , Ni,j ≥ 0 ∀i, j. (15)

A=M
b N
However, our attempts in this direction were unsuccessful, and approximations of A obtained in this way did not provide
useful forecasts.
As {x}ni=1 > 0 element-wise for all i, we also attempted to define:
X
e1 = log(X1 ), X
e2 = log(X2 ), A
e=X e1 )† ,
e 2 (X (16)
and proceeding as:  
xt+1 = exp Ae log (xt ) . (17)
Note the logarithm and exponential functions in (16), (17) refer to the elementwise operations. While the xi obtained in
this manner remained nonnegative, and accurately reconstructed the training period, extrapolating beyond the training
period resulted in oscillatory, nonphysical solutions, even for just a single time-step, making this approach unsuitable
for forecasting, at least for the problems studied herein.
Instead, we formulate nonnegative DMD as the following:
Definition. Let X ∈ Rm×n and let X1 , X2 be snapshot matrices as defined in (11). We define the nonnegative Dynamic
Mode Decomposition as the operator A satisfying:
arg min ∥AX1 − X2 ∥F , Ai,j ≥ 0 ∀i, j. (18)
A

In general, the solutions to (15) and (18) will not coincide. As the definition of the DMD operator (13) is motivated by
dynamics (12), we believe that (18), defined directly in terms of the underlying time dynamics in the data, is the correct
formulation for the applications in the present.
The computation of (18) is nontrivial. Note that well-known algorithms exist to solve the related matrix-vector
nonnegative least squares (NNLS) problem: [54]:
arg min ∥Ax − b∥2 xj ≥ 0 ∀j. (19)
x

Problems related to (18), in which one searches for a nonnegative matrix, have been studied in [55, 56, 57, 58]. We
remark that, in each of these cases, it is assumed that one has access to the matrix A and is attempting to find X1 , that is
solving:
arg min ∥AX1 − X2 ∥F , X1i,j ≥ 0 ∀i, j. (20)
X1
The DMD problem is the opposite, since X1 is known from the data and A is our unknown. However, recalling that
∥M ∥F = ∥M T ∥F [52] for any matrix M , the transposed problem:
arg min ∥X1T AT − X2T ∥F , ATi,j ≥ 0 ∀i, j (21)
AT

is equivalent to (20). Note that:

m
X
∥X1T AT − X2T ∥2F = ∥X1T (Ak,1:m )T − (X2 k,1:(n−1) )T ∥22 (22)
k=1

Let Vec(M ) denote the vector obtained of stacking the successive columns of a matrix M on top of one another. Then
we can write (22) as:
∥X1T AT − X2T ∥2F = ∥(I ⊗ X1T )Vec(AT ) − Vec(X2T )∥22 , (23)
where I is the identity matrix of appropriate dimension, ⊗ denotes the Kronecker product. Note that (23) is a
matrix-vector system. The problem (18) is therefore equivalent to the following matrix-vector NNLS problem:
arg min ∥(I ⊗ X1T )Vec(AT ) − Vec(X2T )∥2 , Vec(AT )j ≥ 0 ∀j. (24)
Vec(AT )

7
 PRIME AI paper

Standard NNLS algorithms (see e.g. [54]) can be applied on (24). Converting Vec(AT ) back into matrix form and
transposing, one obtains a solution to (18). For small problems, this approach is feasible, and it was used throughout
the present.
However, for high-dimensional data, or for datasets with many observations, the direct vectorization approach may be
untenable. In [56], the authors discuss several methods for solving matrix-matrix NNLS problems, including techniques
that are strictly equivalent to (24), as well as techniques that seek sparse solutions to (20) directly. Note that the
vectorized problem (24) has the structure:
 T 
X1 0 ... 0 A(1,:) X2 (1,:)
  
T
 0 X1 ... 0   A(2,:)   X2 (2,:)  2

arg min  .  ..  −  ..  , (25)

.. .. ..     
A  .. . . .  . . 2
0 0 ... X1T A(m,:) X2 (m,:)
and hence can be regarded m independent NNLS problems:
arg min ∥X1T A(k,:) − X2 (k,:) ∥22 , A(k,j) ≥ 0 ∀ j. (26)
A(k,:)

Since X1 , X2 are known, and each A(k,:) can be computed independently, (24) is easily parallelized. We remark that
while the problems (24) and (26) are equivalent in theory, in practice, solutions to NNLS for underdetermined systems
are not unique in general, and must be obtained through iterative procedures. The solutions obtained when solving the
formulations (24) and (26) will therefore differ in practice. In the present, the authors observed that such differences
were trivial; however, a comprehensive investigation of this issue was not pursued.
We remark that related work on a nonnegative DMD variant was pursued in [47]. The formulation introduced in the
present differs from [47] in that we require that the approximated operator A be strictly nonnegative, whereas the ap-
proach in [47] requires only that the dynamic modes, or DMD eigenvectors, are nonnegative. However, as nonnegativity
is not required for the corresponding DMD eigenvalues, the reconstructed DMD operator is not nonnegative in general.
We note that this difference in formulation reflects the primary intended usage of each algorithm, as the approach in [47]
was developed principally for diagnostic purposes and dimension reduction, whereas the present algorithm is intended
for usage in forecasting and extrapolation.

2.3.3 Application to PWDH age structure model

We use the time- and age-dependent mortality curves obtained with the EKI for the years 2009-22 to forecast the
evolution of the PWDH population over the years 2023-30. For the years 2009-22, we solve the model (1), directly
applying the µ(a, t) obtained through the EKI procedure.
To obtain mortality rates for the years 2023-30, we apply the nonnegative DMD algorithm outlined in section 2.3.2
defined as:

M = arg min ∥M [µ2009 µ2010 ... µ2018 ] − [µ2010 µ2011 ... µ2019 ]∥F , M ≥ 0, (27)
M

where the notation M ≥ 0 indicates that M is entry-wise nonnegative. The 2023 mortality rate is then defined as
M µ2022 , with the following years obtained by iterated multiplication by the M . We did not include the years 2020-22
in training the nonnegative DMD operator, as the jump in mortality rate due to COVID-19 likely represents a temporary
jump in mortality, rather than a long-term trend, as indicated by preliminary 2023 mortality data returning to pre-COVID
levels [59].
Similarly, we use nonnegative DMD to project age-structured annual new HIV diagnoses over the years 2023-30 as:
Λ = arg min ∥Λ[λ2009 λ2010 ... λ2018 ] − [λ2010 λ2011 ... λ2019 ]∥F , Λ ≥ 0. (28)
Λ

As with mortality rates, we disregard the years 2020-22 in projecting HIV diagnoses, as available evidence suggests
that the large drop in HIV diagnoses observed in 2020, followed by rebounds in 2021 and 2020, was primarily caused
by disruption and subsequent recovery in HIV testing and diagnosis [60, 61, 62]. However, estimated underlying
HIV incidence continued to follow the decreasing trends observed in the years immediately preceding the COVID-19
pandemic [20]. Therefore, the increases in HIV diagnoses observed in 2021 and 2022 likely reflect a recovery of
diagnoses missed during 2020 due to reduced testing, rather than changes in HIV transmission. Hence, the pre-2019
trends in new HIV diagnoses are likely more reliable for informing future trends in new HIV diagnoses over the coming
years.

8
 PRIME AI paper

2.3.4 Mathematical analysis

We provide a series of results that establish the key properties of nonnegative DMD. Although the results are presented
in some generality, we assume some additional conditions on the underlying data, motivated by the specific problem
studied in the present. Accordingly, they may not apply universally.
Letting X1 , X2 ∈ Rm×n−1 be as in (11), we assume additionally:

• Strict positivity:
(X1 )i,j > 0, (X2 )i,j > 0 ∀ i, j (29)
• Temporal regularity:
(X1 )i,j < 2(X2 )i,j , (X2 )i,j < 2(X1 )i,j (30)
.

Both assumptions hold for the data µ2009−19 , λ2009−19 used for (27), (28), respectively. The meaning of (29) is
straightforward, and a similar assumption was employed in Theorem 1 in Section 2.1.1. Assumption (30) states that
the local temporal variation at each observation point (age in the present document) must remain bounded, and should
not more than double, or less than halve, from time-step to time-step (the factor of two is chosen for convenience).
As DMD assumes underlying continuity in time [51, 53], (30) should hold for any data sufficiently well-resolved in
time; in the absence of noise, it should be possible in principle, to obtain a time series satisfying (30) by increasing
time-resolution of the measurements. However, we recognize that, in practice, time-series data is subject to limitations
and (30) may not hold in general.
Theorem 2. Any M solving (27), or Λ solving (28), is nonzero.
Proof. We provide a result for Problem (27), and note that an identical argument applies to (28). We prove
contrapositive of the theorem statement, and show that if M is a matrix containing only zeroes, it cannot be a solution
of (27). For ease of notation, let:
µ
b 2009−18 = [µ2009 µ2010 ... µ2018 ], µ
b 2010−19 = [µ2010 µ2011 ... µ2019 ], (31)

and proceed by contradiction. Denote as ⃗0 ∈ Rm×m the m × m matrix whose entries are identically zero, and suppose
that ⃗0 solves (27). Then:
∥⃗0b
µ2009−18 − µ
b 2010−19 ∥F = ∥b
µ2010−19 ∥F ≤ ∥Ab
µ2009−18 − µ
b 2010−19 ∥F ∀A ≥ 0. (32)
Let I be the m × m identity matrix. Clearly I ≥ 0. Observe that:
∥I µ
b 2009−18 − µ
b 2010−19 ∥F = ∥b
µ2009−18 − µ
b 2010−19 ∥F . (33)
Squaring the right-hand side of (33) and applying the definition of the Frobenius norm gives:

2018
X 2018
X m
X
∥µj − µj+1 ∥22 = (µi,j − µi,j+1 )2
j=2009 j=2009 i=1
2018
X m
X
= µ2i,j + µ2i,j+1 − 2µi,j µi,j+1
j=2009 i=1
2018
X m
X (34)
< 2µi,j µi,j+1 + µ2i,j+1 − 2µi,j µi,j+1
j=2009 i=1
2018
X m
X
= µ2i,j+1
j=2009 i=1

µ2010−19 ∥2F ,
= ∥b
where the third line follows from the temporal regularity assumption (30). From (34), it follows that:
∥Iµ2009−18 − µ2010−19 ∥F < ∥µ2010−19 ∥F = ∥⃗0 µ2009−18 − µ2010−19 ∥F , (35)
contradicting (32), implying that the zero matrix cannot be a solution of (28), as was to be shown.

9
 PRIME AI paper

Interpretation in terms of the Perron-Frobenius operator. In this portion of the article, we provide an interpretation
of nonnegative DMD in terms of the Perron-Frobenius (P-F) operator. We note that numerical methods for the Perron-
Frobenius operator have been studied in other settings [63, 64, 65, 66]. DMD is generally interpreted as a numerical
approximation of the Koopman operator, the adjoint of the Perron-Frobenius operator [53, 67]. Roughly speaking, the
Koopman operator acts on functions of the system state (commonly called observables), while the P-F operator acts on
densities of the system state.
Since the P-F and Koopman operators are adjoint, one can use either to describe a given dynamical system [68].
Nonetheless, the literature on the numerical approximation of the two operators is somewhat distinct. Literature on
the approximation of the P-F operator tends to focus on Ulam’s method and its variants [63, 65, 66, 69, 70, 71], while
approximation of the Koopman operator focuses on DMD and its variants [67, 53, 46, 40, 72]. Note that this statement
is only a broad, general characterization, and several works discuss the approximation of both operators [63, 73, 74]
and/or hybrid-type approaches incorporating aspects of both DMD and Ulam’s method [75, 76].
In the following, we describe why, in the context of the present work, and problem (27) in particular, the P-F operator
provides a natural interpretation. We remark that a full description of the Perron-Frobenius and Koopman operators is
beyond the scope of the current work, and we recommend that the reader consult [73] for a comprehensive discussion
of the numerical approximation of the P-F and Koopman operators, and [68] for the underlying theory. We note that
this description is not, nor is intended to be, fully rigorous. Rather, it is intended as an intuitive explanation for the
nonnegative DMD formulation used herein, and how it differs from other common DMD approaches within the context
of Koopman/P-F theory.
Consider a measure space defined by the three-tuple M := (A, B, ν) where
A = R+ = [0, ∞), the nonnegative real numbers, is the space of possible ages, B is a σ−algebra on A and ν is
a σ−finite (positive) measure on A, hereafter assumed to be the standard Borel measure unless otherwise specified.
Let S : A → A be a measurable mapping which advances age of the population. For simplicity and without loss of
generality, we assume S advances age forward one unit; hence S moves a member of the population aged a to age
a + 1. For simplicity, we disregard new entries λ(a, t) for the moment.
Let ut (a) ∈ L1 (A, ν) describe the population distribution at time t. We define the support of ut as:
supp(ut ) := {a ∈ A|ut (a) ̸= 0}. (36)
Since human lifespans are finite, we may assume:
ν(supp(ut )) < ∞. (37)
at all t.
Since ut (a) ∈ L1 (A, ν), ut (a) ≥ 0 ∀ a ∈ A, we may also define the population measure u
bt ∈ M:
Z
u
bt (X) = ut (a)da, X ∈ B. (38)
X
Define the three-tuple M
c := (A, B, u bt ) and let µbt (X) ∈ Mc be the mortality measure at a time t, measuring the
number of deaths in the age-set X of the population u at time t. Clearly, if u bt (X) = 0 for some X ∈ B, then
bt (X) = 0 and hence, by the Radon-Nikodym theorem [68], there exists a density function µt ∈ L1 (A, u
µ bt ) such that:
Z Z
µ
bt (X) = µt (a)b
ut (da) = µt (a)ut (a)da, X ∈ B. (39)
X X

From the Riesz Representation Theorem [68], there is a linear operator ψt (u) on the dual space of L1 (A, ν) such that
for any u ∈ L1 (A, ν):
Z ∞ Z ∞
ψt (u) := u(a)b
µt (da) = u(a)µt (a)da = (u, µt ) = µ bt (A), (40)
0 0
where (·, ·) indicates the scalar product.
We now recall that our nonnegative DMD operator A is recovered as the solution of:
arg min ∥X1T AT − X2T ∥F , ATi,j ≥ 0 ∀i, j. (41)
AT
For the sake of concreteness, suppose we are considering the mortality rate-recovery problem (27). Then the matrices
X1 and X2 have the structures:  T   T 
µ2009 µ2010
µT2010  µT2011 
X1T =  .  , X2T =  .  . (42)
   
 ..   .. 
µT2018 µT2019

10
 PRIME AI paper

Intuitively, we may regard the discrete expression µTt u as an analogue of:

Z ∞
µTt u ≈ u(a)bµt (da) = (u, µt ) = ψt (u). (43)
0

Following a similar intuition as used in (43), we note that the expression:

(AX1 )T u = X2T u (44)
is a discrete analogue of:
(u, Aµt ) = (u, µt+1 ) (45)
which, together with (40):
ψt+1 (u) = (u, µt+1 ) = (u, Aµt ) = (A∗ u, µt ) = ψt (A∗ u) = Aψt (u). (46)
Consequently, the nonnegative DMD operator can be interpreted as an approximation of the Perron-Frobenius operator
applied to the mortality measures µ bt (or their related densities concerning the population measure), thereby evolving
them over time [68]. By employing NNLS to compute A, we ensure that the projected µt remain on the positive cone
and in L1 (A, u
bt ) for every t. In other words, non-negative DMD guarantees that densities are transformed into densities,
as desired.

3 Results
3.1 Reconstruction of age-dependent PWDH mortality curves, 2009-22

By applying the Kalman Inversion procedure outlined in the previous section, we observe a strong agreement with the
data in Fig. 2.
This figure illustrates the age distribution of mortality for each year, juxtaposing real data (top panel) with simulated
mortality (bottom panel). The simulated and observed mortality are in agreement both qualitatively and quantitatively.
In Fig. 3, we depict the annual mortality rate per age class over time. The color scale ranges from 0.005 (dark blue)
to 0.04 (red). This visualization indicates a trend of decreasing annual mortality, evidenced by the expanding blue
regions from 2010 to 2022. However, we note that at the beginning of the COVID-19 pandemic (2020), this trend slows
somewhat, and mortality rates increases slightly. The red and yellow regions, which correspond to mortality rates at
older ages, also show a strong diminishing trend in time until the beginning of the COVID-19 pandemic. The results
suggest that the COVID-19 pandemic caused mortality rates across the 55+ age cohorts over the years 2020-21 to revert
approximately to their approximate 2013 levels [5, 77].
In Figure 4, we depict several age-dependent mortality curves for individual years, over the entire age range (left panel),
and focusing on the important 40-80 age range (right panel). Mortality decreases throughout the entire age range;
however, as noted previously, these trends reverse in 2020, and mortality rates over the years 2020-21 increase among
the 55+ cohort to levels similar to 2013-16. The 2022 rates show the effects of COVID subsiding among the 55+ cohort,
with mortality rates consistent with levels observed during the years 2016-19 among PWDH aged 55-75, and at their
lowest rates ever for PWDH aged 75+. However, we note that, even during the peak years of COVID-19 in 2020-21,
mortality among older PWDH was still lower than 2010 levels, which suggests that the effects driving the decreases in
mortality risk among PWDH over the past 15 years were more significant, on net, than the effects of the COVID-19
pandemic on PWDH mortality.

3.2 Forecasting PWDH mortality, new diagnoses and PWDH population age-structure, 2023-30

3.2.1 Future PWDH mortality

In order to forecast the evolution of the mortality rate per age-class, we employ the nonnegative DMD as described in
Sec. 2.3.3. In Fig. 5, we depict the annual age-dependent probability of mortality for the years 2022, 2024, 2026, 2028,
and 2030. The nonnegative DMD algorithm forecasts mortality rates to continue decreasing substantially in the 40-55
age range, with annual mortality projected to decrease 15.6% in this age group from 2022 to 2030. Among PWDH aged
55-75, we also project future decreases in mortality, however these are comparatively modest, decreasing approximately
7.5% from 2022 to 2030. For PWDH aged 75 to 90, mortality is projected to decrease approximately 15.6% from 2022
to 2030.
Analyzing the spectrum of M , we find that it has a unique maximum eigenvalue ξ1 = ρ(M ) = 1 (consistent with the
interpretation of M as an approximation of the Perron-Frobenius operator [73, 68]), with second-largest eigenvalue

11
 PRIME AI paper

Figure 2: The comparison of surveillance (top) and simulated (bottom) mortality.

|ξ2 | = 0.87. We denote the corresponding eigenvectors as v 1 and v 2 , respectively, and express µ2022 as a linear
combination of the eigenvectors v i of M : X
µ2022 = ai v i . (47)
i
Observe that: X
lim M k µ2022 = 1k a1 v 1 + 0.87k a2 v 2 + ξik ai v i = a1 v 1 , (48)
k→∞
i≥3

since |ξi | ≤ .8909 < 1, ∀i ≥ 3.

The eigenvector a1 v 1 of M corresponding to ξ1 = 1 thus gives a long-term projection for the asymptotic age-dependent
PWDH mortality rate, with age-dependent PWDH mortality converging to a1 v 1 like .87t . We plot this eigenvector
as ‘Long-term limit’ in Fig. 5. As the figure shows, beyond 2030, mortality among PWDH is projected to decrease
significantly among PWDH aged 35-55 older than 80, and moderately among PWDH aged 60-80. We note that these
forecasts are based on pre-2019 trends in PWDH mortality and hence cannot account for any changes beyond those
pre-existing trends. As such, these forecasts, especially the asymptotic limit, should be interpreted with caution.
Remark. As M is nonnegative, if it is additionally irreducible, then we can guarantee that v 1 has strictly positive
entries from the Perron-Frobenius theorem. However, establishing irreducibility for a general M is difficult, given the
lack of a closed-form expression. Nonetheless, our experiments over an ensemble of M consistently showed that M
has a unique maximum eigenvalue of 1, with a strictly positive corresponding eigenvector (we note that the spectral gap
varied across the different M ).

12
 PRIME AI paper

3.2.2 Future HIV diagnoses

To forecast future new HIV diagnoses by age, we employed the nonnegative DMD algorithm on the age-depdendent
diagnosis data as in (28). These data were defined using NCHHSTP surveillance data [5]. New HIV diagnoses
are projected to decrease at approximately 3% per year, declining to approximately 28,000 new diagnoses in 2030.
Furthermore, the age structure of new HIV diagnoses is not projected to show significant change in the coming years.
We depict both the number and percentage of new HIV diagnoses by age group in Fig. 6.

3.2.3 PWDH population age structure, 2023-30

Using projected PWDH mortality (27) and projected new HIV diagnoses (28), we solve the system (1) through year-end
2030 to project the evolution of the PWDH age-structure over the years 2023-30.
The total PWDH population is projected to grow over the simulation period, increasing to 1,160,000 by the end of 2030.
As a whole, the PWDH population is also projected to age significantly and rapidly in the coming years. At year-end
2024, an estimated 42.5% of PWDH were 55 and older, 19.4% 65 and older, and 4.2% 75 and older. By year-end 2030,
we project the percentage of PWDH over 55 to increase to 47.% and those over 65 to 26.1%. The percentage of PWDH
over 75 is projected to more than double over a five year span, reaching 8.5%, by the end of 2030. This information is
depicted in Fig. 7.
We remark, however, that not all age groups are projected to increase uniformly. As discussed in the previous subsections,
future PWDH mortality rates are projected to decrease, however, such decreases are highly nonuniform based on age.
Conversely, while new HIV diagnoses are expected to decrease, the projected age distribution does not show significant
change. By 2030, this results in the formation of a bimodal-like population PWDH age distribtution, in which there are
fewer PWDH approximately aged 50 compared to either age 40 and age 60 (Fig. 8).

4 Discussion
In the present work, we have examined age-dependent data on PWDH in the United States to better understand
how the developments over the previous 15 years have changed the age structure of the PWDH population in the
United States and, furthermore, what implications these changes may have going forward. Starting with a simple
age-structured population model, we used an Ensemble Kalman Filter Inversion, together with discrete age-structured
HIV surveillance data, to reconstruct age-dependent PWDH mortality over the years 2008-22. We found that, due to the
widespread availability of effective ART, mortality has declined significantly among PWDH since 2008, particularly
among PWDH aged 40-80 years. While this trend reversed somewhat among PWDH aged 55 and older in 2021-22 due
to the COVID-19 pandemic, data from 2022 suggest that COVID effects have subsided, likely signaling a return to
pre-COVID trends.
We then developed a novel variant of Dynamic Mode Decomposition, nonnegative DMD, to develop projections of
future changes in PWDH mortality and diagnoses. Projections of future PWDH mortality rates suggest that PWDH
mortality will continue to decrease in the coming years, with the largest decreases expected among PWDH aged 40-55
and those older than 80. While mortality rates are expected to decrease among other age groups as well, smaller
decreases are projected. We note that these projections assume a continuation of pre-existing trends, and cannot account
for major changes in the coming years, such as medical breakthroughs.
Our projections of future HIV diagnoses suggest a slow, but notable, decrease in annual HIV diagnoses going forward,
with around 28,000 diagnoses projected in 2030. Interestingly, our methods did not indicate a major change in the age
structure of new HIV diagnoses, which has remained largely static in recent years, and is projected to remain so.
Applying our reconstructed and projected mortality and diagnosis data to our core PWDH age model, we then projected
the age structure of the PWDH population over the coming years. Our simulations suggest that the PWDH population
will age significantly. By 2030, the percentage of PWDH aged 55 years and older is expected to increase significantly
(from 40% to 47.4%). Furthermore, these increases are more dramatic for older age groups; the percentage of PWDH
aged 65 and older is expected to increase to 26.1% by 2030, and the portion of those aged 75 years and older is expected
to more than double by 2030.
We note that the analysis is subject to several important limitations. As previously mentioned, the projections of future
changes in PWDH mortality and diagnoses assume continuation of pre-existing trends. It is possible that new medical
technologies may alter the landscape of HIV prevention and care significantly. This is particularly important for new
HIV diagnoses, as improvements such as rapid expansion of pre-exposure prophylaxis (PrEP) coverage or increased
levels of viral suppression (for instance, due to long-acting injectable ART [78]), may result in fewer new HIV diagnoses
than projected [79]. Furthermore, new HIV diagnoses were modeled as a linear source term, and any dependence on the

13
 PRIME AI paper

Figure 3: Annual probability of mortality at a given age, in time. Note the consistent decrease in time, punctuated by
increases in the upper age ranges in 2020-22 caused by the COVID-19 pandemic.

PWDH age structure is assumed to be captured implicitly via any trends present in the data. However, the relationship
between new the quantity and age distribution of new HIV diagnoses and the current PWDH population state may be
more complex, and future modeling efforts should seek to better explore and define this relationship.
Crucially, the current analysis represents only a first step towards developing comprehensive models for aging in the
PWDH population. Future extensions should include other aspects of HIV prevention and care. For example, beyond
biological age, time since acquiring HIV infection and time on ART may also be important factors in determining risk
for certain commodities. Other extensions, including demographic stratification, may be similarly significant.
On the mathematical end, further analysis of nonnegative DMD is necessary. In the present work, we introduced the
basics of the method and provided some elementary results for the current problem, as well as some intuitive arguments.
Given the application-oriented nature of the current work, we elected to leave further exploration to future research.
However, it is important to properly formalize how nonnegative DMD fits within the larger family of DMD methods.
More rigorous analysis is necessary to establish when certain conditions observed in the current analysis, such as
the existence of a limiting eigenvector with an associated simple eigenvalue of 1, can be expected to hold. Finally,
developing more sophisticated and efficient computational approaches for nonnegative DMD, particularly those that
exploit its parallelizability, are important for extending its application to larger-scale problems.
Our analysis highlights the urgent need to prepare for the aging population of PWDH. Current data suggests that PWDH
are more vulnerable to aging-related comorbidities, requiring that the healthcare system be properly equipped to handle
the potential rapid increase in such health conditions. Furthermore, in the case of preventable age-related comorbidities,
the current analysis also emphasizes the need to develop and implement prevention programs to help reduce potential
burden on the healthcare system as much as possible. Urgency is necessary, as model projections show that the portion
of PWDH aged 75 and older will more than double over a five-year span. Beyond HIV care, using mathematical models
to account for aging populations is an important direction for research in public health, and its importance is likely to
increase in importance in the coming years.

Acknowledgments
E. Iacomini is partially supported by the Italian Research Center on High-Performance Computing, Big Data and
Quantum Computing (ICSC) funded by MUR Missione 4-Next Generation EU (NGEU) [Spoke 1 “Future HPC & Big
Data"]. The authors are members of the INDAM GNCS (Italian National Group of Scientific Calculus. The authors

14
 PRIME AI paper

Figure 4: Mortality curves by age for several years, plotted side-by-side. The left panel shows mortality probability
over the entire age range; the right panel focuses more closely on the important 40-80 age range.

Figure 5: Projection of age-dependent mortality through 2030. The maximum eigenvector is the asymptotic limit of the
projected future age-dependent PWDH mortality.

would like to acknowledge Stefan Klus, Matthew Colbrook, Siobhan O‘Connor and Ruiguang Song for their helpful
input and advice.

References

[1] Adam Trickey, Caroline A Sabin, Greer Burkholder, Heidi Crane, Antonella d’Arminio Monforte, Matthias Egger,
M John Gill, Sophie Grabar, Jodie L Guest, Inma Jarrin, et al. Life expectancy after 2015 of adults with HIV on
long-term antiretroviral therapy in Europe and North America: a collaborative analysis of cohort studies. The
Lancet HIV, 10(5):e295–e307, 2023.
[2] Karin A Bosh. Vital signs: deaths among persons with diagnosed HIV infection, United States, 2010–2018.
MMWR. Morbidity and Mortality Weekly Report, 69, 2020.
[3] Julia L Marcus, Wendy A Leyden, Stacey E Alexeeff, Alexandra N Anderson, Rulin C Hechter, Haihong Hu,
Jennifer O Lam, William J Towner, Qing Yuan, Michael A Horberg, et al. Comparison of overall and comorbidity-
free life expectancy between insured adults with and without HIV infection, 2000-2016. JAMA network open,
3(6):e207954–e207954, 2020.

15
 PRIME AI paper

Figure 6: Projection of age-dependent HIV diagnoses through 2030. We project a slight decrease in overall diagnoses,
with approximately 28,000 new HIV diagnoses in 2030 (compared to 34,500 in 2023).

Figure 7: Projection of the PWDH age structure in the United States through 2030. We observe a gradual aging of the
PWDH population. By 2030, 48.6% of the PWDH population is projected to be over 55 and 27.4% over 65.

16
 PRIME AI paper

Figure 8: Evolution of the PWDH population age structure over the years 2012-2030. By year-end 2030, we see a
bimodal distribution forming, with more PWDH around age 40 and age 60, as compared to PWDH around age 50.

[4] Aysel Gueler, André Moser, Alexandra Calmy, Huldrych F Günthard, Enos Bernasconi, Hansjakob Furrer,
Christoph A Fux, Manuel Battegay, Matthias Cavassini, Pietro Vernazza, et al. Life expectancy in HIV-positive
persons in Switzerland: matched comparison with general population. Aids, 31(3):427–436, 2017.
[5] Centers for Disease Control and Prevention. NCHHSTP AtlasPlus. https://www.cdc.gov/nchhstp/atlas/index.htm,
2024.
[6] M Bloch, M John, D Smith, TA Rasmussen, and E Wright. Managing HIV-associated inflammation and ageing in
the era of modern ART. HIV medicine, 21:2–16, 2020.
[7] Jean Joel Bigna, Aude Laetitia Ndoadoumgue, Jobert Richie Nansseu, Joel Noutakdie Tochie, Ulrich Flore Nyaga,
Jan René Nkeck, Audrey Joyce Foka, Arnaud D Kaze, and Jean Jacques Noubiap. Global burden of hypertension
among people living with HIV in the era of increased life expectancy: a systematic review and meta-analysis.
Journal of Hypertension, 38(9):1659–1668, 2020.
[8] Berta Rodes, Julen Cadinanos, Andres Esteban-Cantos, Javier Rodríguez-Centeno, and José Ramón Arribas.
Ageing with HIV: Challenges and biomarkers. EBioMedicine, 77, 2022.
[9] Rifqah Abeeda Roomaney, Brian van Wyk, and Victoria Pillay-van Wyk. Aging with HIV: increased risk of HIV
comorbidities in older adults. International journal of environmental research and public health, 19(4):2359,
2022.
[10] Anne Marie Fosnacht. Older Adults With HIV: An In-Depth Examination of an Emerging Population: Hauppague,
NY: Nova Publishers, 127 pp., 117 (hardback)., 2013.
[11] Lauren F Collins and Wendy S Armstrong. What it means to age with HIV infection: years gained are not
comorbidity free. JAMA Network Open, 3(6):e208023–e208023, 2020.
[12] Ni Gusti Ayu Nanditha, Adrianna Paiero, Hiwot M Tafessu, Martin St-Jean, Taylor McLinden, Amy C Justice,
Jacek Kopec, Julio SG Montaner, Robert S Hogg, and Viviane D Lima. Excess burden of age-associated
comorbidities among people living with HIV in British Columbia, Canada: a population-based cohort study. BMJ
open, 11(1):e041734, 2021.
[13] Kathy Petoumenos and Signe W Worm. HIV infection, aging and cardiovascular disease: epidemiology and
prevention. Sexual health, 8(4):465–473, 2011.
[14] Robert H Paul, Sarah A Cooley, Paola M Garcia-Egan, and Beau M Ances. Cognitive performance and frailty in
older HIV-positive adults. JAIDS Journal of Acquired Immune Deficiency Syndromes, 79(3):375–380, 2018.
[15] Farrah J Mateen and Edward J Mills. Aging and HIV-related cognitive loss. Jama, 308(4):349–350, 2012.
[16] Lauren Smith, Scott Letendre, Kristine M Erlandson, Qing Ma, Ronald J Ellis, and Shelli F Farhadian. Polyphar-
macy in older adults with HIV infection: Effects on the brain. Journal of the American Geriatrics Society,
70(3):924, 2022.

17
 PRIME AI paper

[17] Giovanni Guaraldi, Manyu Prakash, Christiane Moecklinghoff, Hans-Jürgen Stellbrink, et al. Morbidity in older
HIV-infected patients: impact of long-term antiretroviral use. Aids Rev, 16(2):75–89, 2014.
[18] Anita Chawla, Christina Wang, Cody Patton, Miranda Murray, Yogesh Punekar, Annemiek de Ruiter, and Corklin
Steinhart. A review of long-term toxicity of antiretroviral treatment regimens and implications for an aging
population. Infectious diseases and therapy, 7:183–195, 2018.
[19] Jepchirchir Kiplagat, Dan N Tran, Tristan Barber, Benson Njuguna, Rajesh Vedanthan, Virginia A Triant, and
Sonak D Pastakia. How health systems can adapt to a population ageing with HIV and comorbid disease. The
Lancet HIV, 9(4):e281–e292, 2022.
[20] Centers for Disease Control and Prevention. Estimated HIV incidence and prevalence in the United States,
2018–2022. HIV Surveillance Report, 29(1), 2024.
[21] Monia Puglia and Fabio Voller. HIV/AIDS in Toscana. Technical report, Osservatorio di epidemiologia - Azienda
Regionale di Sanitá della Toscana, 2024.
[22] John Stover and Robert Glaubius. Methods and assumptions for estimating key HIV indicators in the UNAIDS
annual estimates process. JAIDS Journal of Acquired Immune Deficiency Syndromes, 95(1S):e5–e12, 2024.
[23] Keri N Althoff, Cameron Stewart, Elizabeth Humes, Lucas Gerace, Cynthia Boyd, Kelly Gebo, Amy C Justice,
Emily P Hyle, Sally B Coburn, Raynell Lang, et al. The forecasted prevalence of comorbidities and multimorbidity
in people with HIV in the United States through the year 2030: A modeling study. PLoS medicine, 21(1):e1004325,
2024.
[24] Emily P Hyle, Parastu Kasaie, Eli Schwamm, Cameron Stewart, Elizabeth Humes, Krishna P Reddy, Peter F
Rebeiro, Tijana Stanic, Pamela P Pei, Lucas Gerace, et al. A growing number of men who have sex with men
aging with HIV (2021–2031): a comparison of two microsimulation models. The Journal of infectious diseases,
227(3):412–422, 2023.
[25] G. Evensen. Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to
forecast error statistics. J. Geophys. Res, 99:10143–10162, 1994.
[26] G. Evensen and P. J. Van Leeuwen. Assimilation of geosat altimeter data for the agulhas current using the
ensemble Kalman filter with a quasi-geostrophic model. Monthly Weather, 128:85–96, 1996.
[27] S. I. Aanonsen, G. Naevdal, D. S. Oliver, A. C. Reynolds, and B. Valles. The ensemble Kalman filter in reservoir
engineering–a review. SPE J., 14(3):393–412, 2009.
[28] T. Janjic̀, D. McLaughlin, S. E. Cohn, and M. Verlaan. Conservation of mass and preservation of positivity with
ensemble-type Kalman filter algorithms. Monthly Weather Review, 142(2):755–773, 2014.
[29] M. Schwenzer, G. Visconti, M. Ay, T. Bergs, M. Herty, and D. Abel. Identifying trending coefficients with an
ensemble Kalman filter. IFAC-PapersOnLine, 53(2):2292–2298, 2020.
[30] J. Keller, H.-J.H. Franssen, and W. Nowak. Investigating the pilot point ensemble Kalman filter for geostatistical
inversion and data assimilation. Adv. Water Resour., 155, 2021.
[31] Z. Li. An iterative ensemble Kalman method for an inverse scattering problem in acoustics. Modern Physics
Letters B, 34(28):2050312, 2020.
[32] N. B. Kovachki and A. M. Stuart. Ensemble Kalman inversion: a derivative-free technique for machine learning
tasks. Inverse Probl., 35(9):095005, 2019.
[33] A. Yegenoglu, S. Diaz, K. Krajsek, and M. Herty. Ensemble Kalman filter optimizing deep neural networks. In
Conference on Machine Learning, Optimization and Data Science, volume 12514. Springer LNCS Proceedings,
2020.
[34] Lucas Böttcher, Tom Chou, and Maria R D’Orsogna. Forecasting drug-overdose mortality by age in the United
States at the national and county levels. PNAS nexus, 3(2):pgae050, 2024.
[35] Claudia Schillings and Andrew M Stuart. Analysis of the ensemble Kalman filter for inverse problems. SIAM
Journal on Numerical Analysis, 55(3):1264–1290, 2017.
[36] Michael Herty and Giuseppe Visconti. Kinetic methods for inverse problems. Kinetic and Related Models,
12(5):1109–1130, 2019.
[37] Michael Herty, Elisa Iacomini, and Giuseppe Visconti. Recent Trends on Nonlinear Filtering for Inverse Problems.
Communications in Applied and Industrial Mathematics, 13(1):10–20, 2022.
[38] J. Nathan Kutz, Steven L. Brunton, Bingni W. Brunton, and Joshua L. Proctor. Dynamic mode decomposition:
data-driven modeling of complex systems. SIAM, 2016.

18
 PRIME AI paper

[39] Alex Viguerie, Malú Grave, Gabriel F Barros, Guillermo Lorenzo, Alessandro Reali, and Alvaro LGA Coutinho.
Data-driven simulation of Fisher–Kolmogorov tumor growth models using dynamic mode decomposition. Journal
of Biomechanical Engineering, 144(12):121001, 2022.
[40] Joshua L. Proctor and Philip A. Eckhoff. Discovering dynamic patterns from infectious disease data using dynamic
mode decomposition. International Health, 7(2):139–145, 2015.
[41] Gabriel F. Barros, M. Grave, A. Viguerie, A. Reali, and Alvaro L.G.A. Coutinho. Dynamic Mode Decomposition
in Adaptive Mesh Refinement and Coarsening Simulations. Engineering with Computers, 2021.
[42] Alex Viguerie, Gabriel F Barros, Malú Grave, Alessandro Reali, and Alvaro LGA Coutinho. Coupled and
uncoupled dynamic mode decomposition in multi-compartmental systems with applications to epidemiological
and additive manufacturing problems. Computer Methods in Applied Mechanics and Engineering, 391:114600,
2022.
[43] J. Nathan Kutz, Xing Fu, and Steven L. Brunton. Multiresolution dynamic mode decomposition. SIAM Journal
on Applied Dynamical Systems, 15(2):713–735, 2016.
[44] Nicola Fonzi, Steven L. Brunton, and Urban Fasel. Data-driven nonlinear aeroelastic models of morphing wings
for control: Data-driven nonlinear aeroelastic models. Proceedings of the Royal Society A: Mathematical, Physical
and Engineering Sciences, 476(2239), 2020.
[45] Alessandro Alla, Caterina Balzotti, Maya Briani, and Emiliano Cristiani. Understanding mass transfer directions
via data-driven models with application to mobile phone data. SIAM Journal on Applied Dynamical Systems,
19(2):1372–1391, 2020.
[46] Alessandro Alla, Angela Monti, and Ivonne Sgura. Piecewise DMD for oscillatory and Turing spatio-temporal
dynamics. Computers & Mathematics with Applications, 160:108–124, 2024.
[47] Naoya Takeishi, Yoshinobu Kawahara, and Takehisa Yairi. Sparse nonnegative dynamic mode decomposition. In
2017 IEEE International Conference on Image Processing (ICIP), pages 2682–2686. IEEE, 2017.
[48] Margaret A Lampe, Steven R Nesheim, Keydra L Oladapo, Alexander C Ewing, Jeffrey Wiener, and Athena P
Kourtis. Achieving elimination of perinatal HIV in the United States. Pediatrics, 151(5):e2022059604, 2023.
[49] James Dickson Murray and James Dickson Murray. Mathematical biology I: An introduction, volume 18. Springer,
2003.
[50] Xue-Zhi Li, Junyuan Yang, and Maia Martcheva. Age structured epidemic modeling, volume 52. Springer Nature,
2020.
[51] Peter J Schmid. Dynamic mode decomposition of numerical and experimental data. Journal of fluid mechanics,
656:5–28, 2010.
[52] Gene H Golub and Charles F Van Loan. Matrix computations. JHU press, 2013.
[53] J Nathan Kutz, Steven L Brunton, Bingni W Brunton, and Joshua L Proctor. Dynamic mode decomposition:
data-driven modeling of complex systems. SIAM, 2016.
[54] Charles L Lawson and Richard J Hanson. Solving least squares problems. SIAM, 1995.
[55] Mark H Van Benthem and Michael R Keenan. Fast algorithm for the solution of large-scale non-negativity-
constrained least squares problems. Journal of Chemometrics: A Journal of the Chemometrics Society, 18(10):441–
450, 2004.
[56] Nicolas Nadisic, Jeremy E Cohen, Arnaud Vandaele, and Nicolas Gillis. Matrix-wise l0-constrained sparse
nonnegative least squares. Machine Learning, 111(12):4453–4495, 2022.
[57] Hyunsoo Kim and Haesun Park. Sparse non-negative matrix factorizations via alternating non-negativity-
constrained least squares for microarray data analysis. Bioinformatics, 23(12):1495–1502, 2007.
[58] D. Kim, S. Sra, and I. S. Dhillon. A non-monotonic method for large-scale non-negative least squares. Optimization
Methods and Software (OMS), Jan. 2012.
[59] Farida B Ahmad. Mortality in the United States—Provisional Data, 2023. MMWR. Morbidity and Mortality
Weekly Report, 73, 2024.
[60] Alex Viguerie, Ruiguang Song, Anna Satcher Johnson, Cynthia M Lyles, Angela Hernandez, and Paul G Farnham.
Isolating the effect of COVID-19-related disruptions on HIV diagnoses in the United States in 2020. JAIDS
Journal of Acquired Immune Deficiency Syndromes, 92(4):293–299, 2023.
[61] Alex Viguerie, Ruiguang Song, Anna Satcher Johnson, Cynthia M. Lyles, Angela Hernandez, and Paul G. Farnham.
COVID-19-related excess missed HIV diagnoses in the United States in 2021. AIDS, 38(6):907–911, 2024.

19
 PRIME AI paper

[62] Elizabeth A DiNenno. HIV testing before and during the COVID-19 pandemic—United States, 2019–2020.
MMWR. Morbidity and Mortality Weekly Report, 71, 2022.
[63] Stefan Klus, Péter Koltai, and Christof Schütte. On the numerical approximation of the Perron-Frobenius and
Koopman operator. Journal of Computational Dynamics, 3(1):51–79, 2016.
[64] Stefan Klus and Christof Schütte. Towards tensor-based methods for the numerical approximation of the Perron–
Frobenius and Koopman operator. Journal of Computational Dynamics, 3(2):139–161, 2016.
[65] Jiu Ding, Qiang Du, and Tien-Yien Li. High order approximation of the Frobenius-Perron operator. Applied
Mathematics and Computation, 53(2-3):151–171, 1993.
[66] Erik M Bollt and Naratip Santitissadeekorn. Applied and computational measurable dynamics. SIAM, 2013.
[67] Steven L Brunton, Marko Budisic, Eurika Kaiser, and J Nathan Kutz. Modern Koopman Theory for Dynamical
Systems. SIAM Review, 64(2):229–340, 2022.
[68] Andrzej Lasota and Michael C Mackey. Chaos, fractals, and noise: stochastic aspects of dynamics, volume 97.
Springer Science & Business Media, 2013.
[69] Leonardo Ermann and Dima L Shepelyansky. Ulam method and fractal Weyl law for Perron-Frobenius operators.
The European Physical Journal B, 75:299–304, 2010.
[70] Oliver Junge and Péter Koltai. Discretization of the Frobenius–Perron operator using a sparse Haar tensor basis:
the sparse Ulam method. SIAM Journal on Numerical Analysis, 47(5):3464–3485, 2009.
[71] Jiu Ding and Tien Yien Li. Markov finite approximation of Frobenius-Perron operator. Nonlinear Analysis:
Theory, Methods & Applications, 17(8):759–772, 1991.
[72] Matthew J Colbrook, Lorna J Ayton, and Máté Szőke. Residual dynamic mode decomposition: robust and verified
Koopmanism. Journal of Fluid Mechanics, 955:A21, 2023.
[73] Stefan Klus. Data-driven analysis of complex dynamical systems. Habilitation thesis, Freie Universität Berlin,
2020.
[74] Stefan Klus, Feliks Nüske, Péter Koltai, Hao Wu, Ioannis Kevrekidis, Christof Schütte, and Frank Noé. Data-driven
model reduction and transfer operator approximation. Journal of Nonlinear Science, 28:985–1010, 2018.
[75] Debdipta Goswami, Emma Thackray, and Derek A Paley. Constrained Ulam dynamic mode decomposition:
Approximation of the Perron-Frobenius operator for deterministic and stochastic systems. IEEE control systems
letters, 2(4):809–814, 2018.
[76] Matthew J Colbrook. The mpEDMD algorithm for data-driven computations of measure-preserving dynamical
systems. SIAM Journal on Numerical Analysis, 61(3):1585–1608, 2023.
[77] Alex Viguerie, Ruiguang Song, Karin Bosh, Cynthia M Lyles, and Paul G Farnham. Mortality among persons
with HIV in the United States during the COVID-19 pandemic: a population-level analysis. JAIDS Journal of
Acquired Immune Deficiency Syndromes, pages 10–1097, 2022.
[78] Alex Viguerie, Jesse O’Shea, Marie Johnston, Daniel Schreiber, Joella Adams, Laurel Bates, Justin Carrico,
Katherine A Hicks, Cynthia M Lyles, and Paul G Farnham. Impact of increased uptake of long-acting injectable
antiretroviral therapy on HIV incidence and viral suppression in the United States under 2021 FDA guidelines.
AIDS, 10.1097/QAD.0000000000004144, 2025.
[79] Robert Walter Eisinger and Anthony S Fauci. Ending the HIV/AIDS pandemic. Emerging infectious diseases,
24(3):413, 2018.

20