Why Technical Improvement Doesn't Always Show as Survival Improvement

A live case-mix analysis of your own unit, 2012 → 2026.

Generated: 2026-09-21 23:13

Everything we improved — bicarbonate buffer, volumetric control, better water, high-flux dialysers, higher Kt/V, trained nurses, frequent labs — should improve survival. And it does, in the right places. But in real-world dialysis populations the crude mortality number often does not move, and sometimes looks worse. That is not a failure of the technical work. It is what happens when four things move at the same time as the improvements do.

The clinical reasoning is correct. The clinical improvements are real. What changed is who we dialyse, when we start them, and how much follow-up each era has actually had — and those three things dominate the crude number.

Your unit's case-mix shift, era by era (live data)

Era	New starters	Mean age at start	% aged <40	% aged 70+	Deaths recorded	Deaths < 1 year
2012–2015	116	55.5	17.2%	22.4%	74	18
2016–2019	134	56.5	14.2%	19.4%	89	24
2020–2023	203	55.2	19.7%	19.7%	84	53
2024–now	82	59.1	12.2%	32.9%	23	16

Change in mean age at start (earliest → newest era)

+3.6 years

A rise of 3.6 years at the starting line is enough to fully mask a 10–20% improvement in relative survival.

Change in new starters per era

-34 patients

A rising intake in an era with a shorter follow-up window will always look like higher mortality, even at identical risk.

The four mechanisms behind the gap

1. Case-mix shift — the dominant effect

In 2012 the unit probably dialysed a younger, less comorbid, more selected population. Since then acceptance criteria have widened everywhere: older patients accepted, more diabetics, more heart failure, more cancer survivors, more frail patients, more emergency starts. The net effect is that the average patient in 2024 is sicker at the starting line than in 2012. Technical improvements of 10–20% in relative risk can be entirely cancelled by a 5-year rise in mean age plus a 10-point rise in diabetes prevalence.

How to test it in your data: Compare survival within the same age band and diabetes status. If 60–69-year-old non-diabetics starting in 2019–2023 survive longer than 60–69-year-old non-diabetics starting in 2012–2015, the technical improvements are working. If not, the gains are absorbed by something else in that group.

2. Improvements mostly affect short-term outcomes

Bicarbonate buffer, volumetric control, better water, high-flux dialysers, higher Kt/V — each of these improves intradialytic symptoms, anaemia, inflammation and short-term survival. But the two things that dominate 5-year mortality in dialysis patients are cardiovascular disease and infection, neither of which is strongly moved by Kt/V, buffer type or dialyser flux. They are moved by access type, blood pressure control, dry-weight accuracy, mineral-bone disease, inflammation, nutrition, and vintage itself.

How to test it in your data: If 1-year survival is improving but 5-year survival is not, that is the expected pattern when the technical improvements you listed are working — the ceiling is elsewhere.

3. Competing risks and the survivor effect

In the earliest era (2012–2015) the patients still alive today are the strongest ones — they have survived 10+ years. A naïve "average survival among deceased" calculation for the old era is therefore averaging over a mixed group, while a recent era that still has many patients alive looks artificially short. This alone will make the old era look better even if nothing clinically changed.

How to test it in your data: Use Kaplan–Meier curves, not means. Compare landmark survival at 1, 2 and 5 years, only counting patients who had that much follow-up.

4. Prevention gap — the incidence side

If new-patient intake is rising, especially in younger patients, more people are reaching dialysis who should not have. This is not a dialysis-quality problem — it is an upstream prevention problem. No amount of high-flux dialysis can compensate for a 45-year-old diabetic reaching ESKD.

How to test it in your data: Track new starters per year and their age at start. If intake is rising and mean age is falling, the case-mix problem is upstream.

What each improvement actually moves

Improvement	What it actually improves	Effect on long-term mortality
Bicarbonate buffer vs acetate	Intradialytic symptoms, acidosis, Hb	Small on 5-year mortality
Volumetric UF control	Fluid accuracy, hypotension	Small–moderate
Water treatment quality	Inflammation, CRP, anaemia	Moderate
High-flux dialysers	Middle molecules, β2-microglobulin	Small in non-diabetics
Higher achieved Kt/V	Urea clearance, short-term survival	Modest, ceiling above 1.4
Trained nursing staff	Access survival, infection	Moderate
Frequent laboratory testing	Earlier detection of problems	Small
AVF-first access policy	Infection, thrombosis, dose	Large on long-term mortality
BP control & dry-weight accuracy	LVH, heart failure, stroke	Large on long-term mortality

The two interventions that move long-term mortality most — access type and blood pressure / dry-weight accuracy — are the ones that changed the least in most units over the same period.

What you can honestly claim

Better anaemia control (higher mean Hb, less ESA use, less transfusion)
Higher achieved Kt/V
Fewer intradialytic symptoms and hypotensive episodes
Fewer catheter-related infections and fewer catheters overall
Better phosphorus and PTH control
Better patient-reported quality of life
Higher 1-year survival in matched subgroups

What you cannot honestly claim without case-mix adjustment

That overall crude survival improved (without case-mix adjustment)
That "average life on dialysis" is longer for recent starters
That newer patients live longer than old-timers (by raw means)

How to prove the improvements are real — three analyses

1. Matched-subgroup survival

Restrict the era comparison to a matched subgroup: age 40–70, non-diabetic, first start. If survival in that group improves across eras, your technical improvements are working.

Output: one line per era for the matched group.

2. Process quality markers

Show the technical work directly: mean Kt/V, % with Kt/V ≥ 1.2, % AVF at initiation, mean Hb, % phosphorus in range, CRBSI per 1000 catheter-days, % planned starts. Each plotted per year.

Output: a "quality improvement" page that responds quickly and reflects the real work.

3. Cause-specific mortality by era

Split deaths by cause per era: cardiovascular, infection, malignancy, withdrawal, unknown. If cardiovascular deaths fall but withdrawal deaths rise, that is the story of an ageing population — not a care failure.

Needs a cause_of_death column.

What to present to administrators, funders, and the team

Show process markers improved year over year. This is the evidence of the work actually done.
Show survival improved in a matched subgroup. This isolates the technical effect from the case-mix effect.
Show cause-specific mortality shifted as expected. Cardiovascular down, withdrawal up — the ageing population effect.
State plainly that the crude mortality reflects a sicker accepted population, not worse care.
Report catheter rate, planned-start rate, and CRBSI rate as quality indicators in their own right — they respond quickly and are not affected by the case-mix trap.

The honest summary

Your clinical reasoning is correct — better dialysis should improve survival. But in real-world dialysis populations, case-mix shifts and the widening of acceptance criteria usually cancel out the technical gains at the population level, while those same gains are clearly visible in process markers and in matched subgroups.

The honest answer to "did our care improve?" is: yes, it improved, and you can prove it — just not by the crude mean-survival number. You prove it by process markers, by matched-subgroup survival, and by cause-specific mortality patterns. That is the correct clinical narrative, and it is the one that protects the unit's credibility when someone looks at the raw numbers and says "survival isn't improving."

✍️ Author: Ahmed Mohmad Rashyd Musleh Registered Staff Nurse