Chronic kidney disease: a quiet loss of filtration, and how it is slowed

Start here: what CKD is

Each kidney is a filter built out of roughly a million microscopic units called nephrons. Blood arrives under pressure, a tuft of capillaries called the glomerulus strains fluid out of it, and the tubule downstream reclaims what the body needs and sends the rest out as urine. Chronic kidney disease (CKD) is the state in which that filtering apparatus is persistently damaged or persistently underperforming — formally, an abnormality of kidney structure or function lasting more than three months, marked either by a filtration rate below 60 mL/min/1.73 m² or by protein leaking into the urine ^[1].

Two features make CKD unusual as a public-health problem. The first is how common it is: pooled population studies put global prevalence at roughly 11–13%, with the large majority sitting in the middle stages rather than at the severe end ^{[2] [1]}. The second is how quiet it is. Kidneys have enough reserve that a person can lose a substantial fraction of function without feeling different, so CKD is typically asymptomatic until late ^[2]. The consequence is a large diagnostic gap — among people with early CKD, fewer than 5% report being aware they have it ^[1]. This is why kidney disease is found by a blood test and a urine test rather than by symptoms, and why professional bodies have argued specifically for earlier identification as a policy goal rather than a clinical nicety ^{[3] [4]}.

In high-income settings CKD is most often attributed to diabetes and hypertension — the same two conditions that dominate cardiometabolic medicine generally ^[1]. CKD is in that sense a downstream organ consequence, which is why it is increasingly framed together with heart and metabolic disease as a single interacting syndrome rather than three separate specialties ^{[5] [6]}.

Why it matters runs in two directions. Forward, CKD can progress to kidney failure, where filtration is no longer sufficient to sustain life and treatment means dialysis or transplantation; globally CKD accounted for about 1.2 million deaths in 2017, and the all-age mortality rate rose 41.5% between 1990 and 2017 ^[7]. Sideways — and for most patients, sooner — CKD is a powerful cardiovascular risk state. Reduced filtration and urinary protein each independently predict cardiovascular events, so most people with CKD are far more likely to have a heart attack, stroke, or heart failure than to ever reach dialysis ^{[8] [9] [10]}.

This review follows three pillars: measurements show what is happening, medicines change the biology, and progress asks what is coming next. Between the second and third sits a simple simulatable model of decline that explains what all the therapies are actually doing.

Pillar 1: measurements

CKD is defined by two numbers, not one. That two-axis structure is the single most useful thing a non-specialist can learn about kidney testing ^[11].

eGFR — how fast the filters work. Glomerular filtration rate is the volume of fluid the glomeruli strain out per minute, normalized to body surface area (mL/min/1.73 m²). Measuring it directly requires a clearance study, so in practice it is estimated from a blood marker. Creatinine is a waste product of muscle metabolism produced at a roughly steady rate and cleared by the kidney; when filtration falls, creatinine backs up in the blood. An estimating equation converts serum creatinine plus age and sex into an eGFR ^[11]. That inference is indirect, and its weakness is exactly the assumption it rests on: creatinine production depends on muscle mass and diet, so extremes of either bias the estimate. In those cases cystatin C, a protein produced by nearly all nucleated cells and less tied to muscle, gives a second and partly independent estimate; combining both markers is the most accurate routine approach ^{[11] [12]}.

The equations themselves changed recently for reasons that were as much conceptual as technical. Older eGFR equations included a race coefficient; the 2021 CKD-EPI refit removed it on the grounds that race is a social rather than biologic construct, and showed that the creatinine–cystatin C equation without race was more accurate than creatinine alone and narrowed the differences between racial groups ^[12].

By eGFR, CKD is graded G1 through G5: G1 (≥90) and G2 (60–89) are normal-or-high and mildly reduced filtration, which only count as CKD when there is other evidence of kidney damage; G3a (45–59) and G3b (30–44) are moderately reduced; G4 (15–29) is severely reduced; and G5 (<15) is kidney failure ^{[11] [2]}. Trial and registry work has since had to define kidney-failure endpoints precisely enough to compare studies, which is why "kidney failure" now carries consensus operational definitions rather than a single threshold used loosely ^[13].

Albuminuria — whether the filter leaks. A healthy glomerulus retains albumin, the most abundant blood protein. When the filtration barrier is damaged, albumin escapes into the urine, which makes it an early damage signal that often appears while eGFR still looks acceptable. It is quantified as the urine albumin-to-creatinine ratio (ACR) on an untimed ("spot") urine sample, using urine creatinine to correct for how dilute the sample is ^[11]. The grades are A1 (<30 mg/g, normal to mildly increased), A2 (30–300, moderately increased), and A3 (>300, severely increased) ^{[11] [2]}. Where only a dipstick or a total protein-to-creatinine ratio is available, published conversions let those be mapped onto the ACR scale, though with real loss of precision ^[14].

The KDIGO heat map. Because the two axes carry partly independent information, staging is not a line but a grid: G1–G5 down one side, A1–A3 across the other, with each cell shaded by risk. The practical meaning is that a person with preserved eGFR and heavy albuminuria can sit in a higher-risk cell than someone with moderately reduced eGFR and no protein leak — so eGFR alone systematically misclassifies risk ^{[11] [10]}. The grid is prognostic, not just descriptive: the eGFR/albuminuria combination predicts cardiovascular outcomes as well as kidney outcomes ^[8].

Albuminuria is also a movable number, not only a static risk label. Analyses across randomized trials found that treatment-induced reductions in albuminuria track treatment effects on kidney disease progression, and observational consortium data show that change in albuminuria predicts subsequent end-stage kidney disease risk — which is what makes it usable as an early readout of whether a therapy is working ^{[15] [16]}.

One live disagreement is worth flagging, because it affects who is even called a patient. Since eGFR declines with normal aging, a fixed 60 mL/min/1.73 m² threshold labels many older adults with stable, non-progressive kidney function as diseased; an age-adapted definition has been proposed to address this, and it remains contested ^[17].

Centerpiece: a simple simulatable model of decline

Everything in the measurement section is a snapshot. What actually determines a patient's future is the rate at which eGFR falls, and there is a deliberately simple model for that.

Treat eGFR as declining approximately linearly in time:

eGFR(t) = eGFR₀ − s · t

where eGFR₀ is the current filtration rate in mL/min/1.73 m², t is time in years, and s is the annual eGFR slope in mL/min/1.73 m² per year. Setting eGFR(t) equal to the kidney-failure threshold of about 15 and solving gives the time to kidney failure:

T_failure = (eGFR₀ − 15) / s

This linear-slope abstraction is not an arbitrary teaching device. It is the quantity the nephrology trials community formally evaluated as a surrogate endpoint: an individual-participant meta-analysis of 14 cohorts (3.76 million participants with baseline eGFR ≥60 and 122,664 with eGFR <60) found that a slower eGFR slope was significantly associated with lower risk of end-stage kidney disease across populations, including those with better baseline kidney function, with the relationship strongest for the 3-year slope ^[18]. A subsequent trial-level meta-analysis extended the same logic to treatment effects, establishing GFR slope as a surrogate endpoint for kidney failure ^[19]. The model form — a per-year rate of decline that maps onto eventual failure — is therefore the grounded object, borrowed here from the endpoint literature.

Parameters. eGFR₀ comes from the patient's current laboratory value. The slope s is the parameter therapies act on. The specific slope values used below are illustrative choices, selected to span the range the trial literature discusses rather than quoted from any single study: a rapidly progressive untreated course at s = 4.0, and a treated course at s = 2.5. Worked out from eGFR₀ = 45 (stage G3a):

• untreated, s = 4.0 → T_failure = (45 − 15)/4.0 = 7.5 years
• treated, s = 2.5 → T_failure = (45 − 15)/2.5 = 12.0 years

Same starting point, same threshold; a reduction in slope buys 4.5 years. This is the entire logic of modern CKD therapy in one line — the drugs in the next section are not repairing nephrons, they are bending the slope.

Assumptions and limits. Real eGFR trajectories are not perfectly straight: some patients progress in steps punctuated by acute kidney injury, some are stable for years, and several therapies produce an initial dip in eGFR before the long-run slope improves, so a short-window slope can invert the sign of the long-run benefit. The surrogate work is explicit that the association is strongest over longer horizons and in rapidly progressive disease, which is precisely where the linear approximation is least strained ^{[18] [19]}. Trial reporting has adapted by separating a "total" slope from a "chronic" slope measured after the acute dip ^[21].

Pillar 2: medicines

The therapeutic story of CKD is three eras stacked on top of each other, and — importantly — they are additive rather than sequential replacements.

Era 1: RAAS blockade. The renin–angiotensin–aldosterone system is the body's blood-pressure and salt-balance controller. Renin from the kidney initiates a cascade that produces angiotensin II, which constricts vessels and stimulates aldosterone release, driving sodium retention ^[22]. Its relevance to CKD is that angiotensin II preferentially constricts the efferent arteriole leaving the glomerulus, raising the pressure inside the filtering tuft; sustained intraglomerular hypertension damages the barrier and drives protein leak. RAAS activity also has direct pro-fibrotic and pro-inflammatory effects on kidney tissue that are separate from its blood-pressure action ^[23]. ACE inhibitors block conversion to angiotensin II; angiotensin receptor blockers (ARBs) block its receptor. Either way, efferent tone falls, intraglomerular pressure falls, and albuminuria falls.

The landmark demonstration that this was kidney protection and not merely blood-pressure lowering came from trials designed to control for pressure. In 1,715 hypertensive patients with type 2 diabetic nephropathy treated to the same blood-pressure target, irbesartan reduced the composite of doubling of serum creatinine, end-stage renal disease, or death by 20% versus placebo and 23% versus amlodipine, with a 33% lower risk of creatinine doubling than placebo — a benefit the calcium-channel blocker arm, at equivalent blood pressure, did not produce ^[24]. Systematic review across renal-outcome trials supported the same conclusion for RAS inhibitors as a class ^[25]. RAAS blockade in proteinuric CKD has been standard of care ever since, and remains the backbone on which everything later was added ^{[26] [1]}.

The obvious follow-up — block the system harder by combining agents — was tried and did not pay off; adding an ARB or a steroidal mineralocorticoid antagonist on top of maximal ACE inhibition ran into adverse-effect ceilings, particularly hyperkalemia and acute kidney injury ^[27]. That failure is what made the next two eras significant: progress had to come from a different mechanism, not a bigger dose of the same one.

Era 2: the SGLT2-inhibitor revolution. Sodium–glucose cotransporter 2 sits in the proximal tubule and reabsorbs most of the glucose the glomerulus filters, along with sodium. Blocking it was designed as a diabetes drug — dump glucose in the urine, lower blood sugar. The kidney effect turned out to be larger and mechanistically distinct. In diabetes, the glomerulus is often hyperfiltering, running at damagingly high pressure ^[28]; SGLT2 blockade delivers more sodium to the distal sensor (the macula densa), which restores tubuloglomerular feedback and constricts the afferent arteriole, dropping intraglomerular pressure by a route entirely different from RAAS blockade ^[29].

The signal appeared first as a secondary finding in cardiovascular safety trials: empagliflozin slowed progression of kidney disease in type 2 diabetes ^[30], and canagliflozin showed the same in the CANVAS program ^[31]. CREDENCE then tested it as a primary kidney question in diabetic nephropathy and confirmed it ^[32]. The two trials that generalized it beyond diabetes are the ones that changed the field:

• DAPA-CKD randomized 4,304 patients with CKD (eGFR 25–75, ACR 200–5000), with or without type 2 diabetes, to dapagliflozin or placebo. It was stopped early for efficacy. Over a median 2.4 years the primary composite (≥50% sustained eGFR decline, end-stage kidney disease, or renal/cardiovascular death) occurred in 9.2% versus 14.5% — hazard ratio 0.61, number needed to treat 19 ^[33]. A prespecified analysis confirmed the effect showed up as a reduced rate of kidney-function decline — the slope, in the language of the model above — in patients with and without diabetes ^[34].
• EMPA-KIDNEY widened the population further: 6,609 patients across a broad range of CKD, including lower albuminuria and lower eGFR than prior trials. Over a median 2.0 years, kidney-disease progression or cardiovascular death occurred in 13.1% versus 16.9% (hazard ratio 0.72), consistent in patients with and without diabetes and across eGFR subgroups ^{[35] [20]}.

Collaborative meta-analyses across the SGLT2 program have since confirmed kidney protection with and without diabetes and quantified it at class level ^{[36] [37] [38]}. The shift this represents is worth stating plainly: a glucose-lowering drug became a kidney drug, then became a kidney drug for people without diabetes ^[39].

Era 3: nonsteroidal mineralocorticoid receptor antagonism. Aldosterone acting on the mineralocorticoid receptor drives inflammation and fibrosis in the kidney beyond its salt-handling role, and the receptor stays inappropriately activated even in patients already on RAAS blockade ^[40]. Older steroidal MRAs (spironolactone, eplerenone) address this but bring hyperkalemia and off-target steroid effects. Finerenone is a nonsteroidal, selective MRA with a different tissue distribution and potency profile; early work showed dose-dependent albuminuria reduction in diabetic nephropathy on top of RAAS blockade ^[41].

Two large outcome trials followed, both enrolling patients already on maximally tolerated RAAS blockade:

• FIDELIO-DKD (5,674 patients with CKD and type 2 diabetes, predominantly stage 3–4 with severely elevated albuminuria) reported the kidney composite — kidney failure, sustained ≥40% eGFR decline, or renal death — in 17.8% versus 21.1% over a median 2.6 years, hazard ratio 0.82 ^[42].
• FIGARO-DKD (7,437 patients across a wider CKD range, including stages 1–2) was powered on cardiovascular outcomes and found 12.4% versus 14.2% over a median 3.4 years, hazard ratio 0.87, driven largely by fewer heart-failure hospitalizations ^[43].

The FIDELITY prespecified pooled analysis of both trials established the combined cardiorenal effect across the full spectrum enrolled ^{[44] [45] [46]}. The trade-off is real and quantified: hyperkalemia occurred more often with finerenone and drove more treatment discontinuation, so potassium monitoring is part of the therapy rather than an afterthought ^[47].

Guideline bodies have absorbed all three eras into a layered regimen — RAAS blockade, plus an SGLT2 inhibitor, plus a nonsteroidal MRA in diabetic CKD with residual albuminuria — codified in the KDIGO diabetes-in-CKD guideline and the joint ADA/KDIGO consensus report ^{[48] [49] [50]}.

Pillar 3: progress

Better biomarkers than creatinine and albumin. Both current measurements are indirect: creatinine reports filtration only after function is already lost, and albuminuria reports barrier damage without saying which mechanism caused it. Candidate panels aim at tubular injury, fibrosis, and inflammation directly ^{[51] [52]}. The clearest proof-of-concept for acting on a biomarker rather than a lab threshold is the urinary-proteomics approach, where a proteomic classifier was used to identify people with diabetes at high risk before overt disease and to trigger early intervention ^{[53] [54]}.

Genetics and precision nephrology. Genome-wide work has identified variants associated with rapid kidney-function decline, moving progression risk partly upstream of clinical measurement ^[55]. The best-developed example is APOL1: risk variants common in populations of West African ancestry confer substantially elevated risk of nondiabetic kidney disease, mechanism has been demonstrated experimentally in podocyte models, and the genetics is now being translated toward clinical use and targeted therapy ^{[56] [57]}. This is also the clearest illustration of why removing race from eGFR equations and taking genotype seriously are complementary rather than contradictory moves ^[12].

AI on trajectories rather than snapshots. Because the model that matters is a slope, kidney disease is a natural fit for methods that learn from longitudinal records. Machine-learning models trained on large clinical datasets have been used to predict diabetic-kidney-disease progression ^[58], and risk scores combining biomarkers with electronic patient data have been derived and validated for predicting progression ^[59]. The methodological caution is well established rather than hypothetical: such models require transparent reporting and external validation before clinical claims, and comparisons of machine learning against conventional regression in kidney outcomes have often found narrower margins than headline claims suggest ^{[60] [61]}.

New agents and combinations. The GLP-1 receptor agonists are the next class arriving with kidney endpoints: FLOW was designed specifically as a kidney-outcomes trial of once-weekly semaglutide in type 2 diabetes and CKD ^[62], semaglutide showed albuminuria and kidney-function effects in overweight and obese populations ^[63], and long-term kidney outcomes were reported from the SELECT cardiovascular trial ^[64]. Whether the three modern classes stack is being tested directly, as in the finerenone-plus-empagliflozin combination trial ^{[65] [66]}, and modeling work has estimated the lifetime cardiovascular, kidney, and mortality benefit of combining SGLT2 inhibitors, GLP-1 receptor agonists, and nonsteroidal MRAs ^{[67] [68]}. Further out, the target is the common final pathway: kidney fibrosis itself, where mechanism-to-medicine programs are still preclinical or early ^{[69] [70] [71]}.