Diabetes follow-up, before patients are lost

Every sugar test done at Anubhav Life Care, Barasat, joined into one timeline per person, checked against follow-up guidelines, and turned into a ranked call list for the care team. Companion to the anaemia and TB dashboards, but longitudinal: the same person, followed across months.

Synthetic cohort shaped like the AKTIV research mirror · 7,888 people · 3.5 years to 23 Sep 2026 · no real patient data. The pipeline reads plain lab rows and runs unchanged on the live mirror.

3,658

people with a diabetic-range result

50.1%

of them lost to follow-up now

1,834 people, 180+ days past due

814

overdue, still reachable

0 to 180 days past due

226

critical + high this week

the top of the call list

One person, many PATIENT_KEYs

AKTIV issues a fresh PATIENT_KEY every month, so a patient tested in January, April and August looks like three strangers. Before any follow-up can be tracked, visits have to be joined back into one person. The mobile number does most of the work; names settle the rest.

14,410 bills14,334 PATIENT_KEYs7,888 people after linking

14,410

bills with a sugar test

14,334

PATIENT_KEYs

almost one per bill

7,888

people

2,913 with 2+ visits, 3.2 keys each on average

Worked example: Subhajit Pramanik

7 visits · 7 PATIENT_KEYs · 7 spellings · one person

Visit	PATIENT_KEY	Name as typed	Phone as typed	Joined by
17 Dec 2024	415744	SUBHAJIT PRAMANIK	96xxx xx616 (patient's own)	Mobile + name
19 Jan 2025	416160	Subhajit Pramanik	96xxx xx616 (patient's own)	Mobile + name
23 Apr 2025	417306	Shubhajit Pramanik	96xxx xx616 (patient's own)	Mobile + name
7 Sep 2025	418994	Subhaijt Pramanik	66xxx xx102 (agent / centre number)	Name, sex, age
25 Nov 2025	420071	SUBHAJIT PRAMANICK	96xxx xx616 (patient's own)	Mobile + name
15 Mar 2026	421538	SUVAJIT PRAMANIK	96xxx xx616 (patient's own)	Mobile + name
18 Jul 2026	423230	Subhajit	96xxx xx616 (patient's own)	Mobile + name

Visits where the phone field held an agent’s or centre’s number, junk, or nothing at all could not be joined on phone. They were attached because exactly one phone-linked person had a matching name, the same sex and a birth year within 2 years.

The rules, in order

as practised at the front desk, now automated

Clean the number. Strip +91 and leading zeros. Blank, too short, or junk such as 9999999999 and 1234567890 counts as no number.
Spot shared numbers. A number carrying more than 15 different people is a collection agent’s, camp desk’s or partner centre’s own phone, not the patient’s. Found 15, each with 67 to 106 people.
Same number, then fuzzy name. On one number, visits join when the names sound alike (Mousumi / Moushumi, Mondal / Mandal, Md / Mohammad), sex agrees and birth years are within 3. A different name on the same number is a family member. A one-digit phone typo with the same name also joins.
No usable number: go by name. Attach to a phone-linked person only if exactly one matches on name, sex and age; otherwise link on name alone, and never on a first name alone.

What is in the phone field

14,410 bills

Patient's own mobile72.9% 10,509 bills
Agent, camp or centre number15.2% 2,190
Blank8.3% 1,194
Junk (9999999999, 12345, NA)3.6% 517

The 15 shared numbers matched the collection roster: 3 camp desks, 8 home-collection agents, 4 partner centres. They are useless for linking, but useful for reaching people: the agent who typed their own number knows the patient.

One number, three people

household mobile 93xxx xx953

Pratima BasuF · 42y · 2 visits
Pratima BoseF · 42y · 2 visits
Tarun BoseM · 50y · 2 visits

1,446 numbers are shared by a household. Treating the number as the patient would give these three people one blended HbA1c history.

How people were linked

7,888 people

Linked on mobile + name6,055
Linked on name, sex, age only324
Single visit, no usable number1,509

1,536 visits with no usable number were attached to a phone-linked person by name. Single-visit records stay separate until a later visit gives them a number.

Is the linking right?

scored against the generator's hidden person id

Join visits on	People	Precision	Recall	Timelines exact
PATIENT_KEY	14,334	100.0%	0.5%	0.8%
Phone number only	6,220	5.1%	59.0%	20.3%
Exact name + sex	4,700	25.8%	65.1%	7.7%
Phone, then fuzzy name (ours)	7,888	91.9%	87.4%	73.8%

PATIENT_KEY almost never joins two visits. Phone alone merges everyone a camp desk registered into one “patient”. Phone then fuzzy name keeps 91.9% of joins correct while recovering 87.4% of true repeat visits. The misses are mostly namesakes with no number, which the resolver refuses to guess.

Plausibility check

HbA1c jumps that biology does not allow

timelines flagged for a human to check

68%

of flagged were really two people

HbA1c reflects three months of sugar, so it cannot swing 2.5 points in a few weeks or go from normal to 8+ without a middle step. When a linked timeline does that, the link is the likely culprit, usually two namesakes with no number. The call list marks these rows so the coordinator confirms identity on the call.

Where the cohort stands today

Everyone with at least one sugar test (HbA1c, fasting, post-meal or random glucose), using each person's latest results after linking.

7,888 peopleas of 23 Sep 2026

Who is in the cohort

sex as recorded; age at latest visit

7,888

people

50 : 50

female : male

425

18–29

1,060

30–39

2,012

40–49

2,073

50–59

1,389

60–69

929

70+

Latest HbA1c

5,460 people with an HbA1c on file

Normal (< 5.7)1,31624%
Prediabetes (5.7 to 6.4)1,34125%
Diabetes (6.5 to 7.9)1,52528%
Diabetes, high (≥ 8.0)1,27823%

Diabetic-range result, by sex and age

% of people with any HbA1c ≥ 6.5, FBS ≥ 126 or PP/random ≥ 200

age

18–2930–3940–4950–5960–6970+

Female

14%

26%

43%

53%

55%

53%

Male

12%

31%

43%

59%

57%

56%

This is a lab cohort, not a population survey: people come here because a doctor suspected diabetes, so prevalence runs far above community figures. The age gradient is the useful part: by the fifties, more than half of those tested are in the diabetic range.

Control among people with diabetes

latest HbA1c, n = 3,173

under 7 (at goal)28% 890
7 to 7.931.7% 1,005
8 to 8.921.3% 676
9 to 9.910.9% 345
10 and over8.1% 257

Complications already showing

people with diabetes, latest result where tested

eGFR under 607.8% 174 / 2,229
Urine ACR 30 or more34% 377 / 1,109
LDL 100 or more67.5% 1,336 / 1,978
Anaemia (Hb < 12 F, < 13 M)44.3% 870 / 1,963

Denominators differ because not everyone was tested, which is its own finding: see the next chart.

Yearly checks actually done

people with diabetes seen in the last 12 months, n = 1,014

2+ HbA1c in the year44.3% 449
eGFR (creatinine)55.3% 561
Urine ACR25.6% 260
Lipid profile48.8% 495

Guidelines ask for all four every year. Even among people who are still coming in, urine ACR, the earliest kidney warning, is done for about one in four.

Following the same people over time

Only possible once visits are linked. Clock starts at each person’s first diabetic-range result. Someone is lost to follow-up once they are more than 180 days past their due date with no sugar test here.

3,658 people with a diabetic-range result3.5 years of bills

The follow-up cascade

people whose first diabetic-range result was at least 12 months ago, n = 2,568; each step counts people who also passed every step above

Diabetic-range result100% 2,568
Came back for any sugar test66.3% 1,702
Repeat HbA1c within 6 months41.4% 1,063
Still testing in the last 12 months22.8% 586
Latest HbA1c under 78.5% 217

Show as table

Step	People	% of start
Diabetic-range result	2,568	100%
Came back for any sugar test	1,702	66.3%
Repeat HbA1c within 6 months	1,063	41.4%
Still testing in the last 12 months	586	22.8%
Latest HbA1c under 7	217	8.5%

Still in care, months after diagnosis

Kaplan-Meier, censored at 23 Sep 2026

Own mobile on file (n = 2,894)
Agent's number or none (n = 764)

Nobody can count as lost before month 9: a 3-month recall plus the 6-month grace. After that the two groups split fast. At 12 months 56% of people with their own number are still in care, against 21% without one.

Show as table

Month	Own mobile: in care	at risk	No own number: in care	at risk
0	100%	2,894	100%	763
6	100%	2,472	100%	609
9	72%	1,655	41%	223
12	56%	1,145	21%	107
18	35%	592	11%	36
24	22%	278	4%	10

Gap between consecutive HbA1c tests

4,209 repeat tests in people with diabetes

34%

≤ 3 mo

38%

3–6 mo

12%

6–9 mo

9–12 mo

12–18 mo

> 18 mo

On schedule (every 3 to 6 months)
Late

Even among people who do come back, 28% of repeat HbA1c tests arrive more than 6 months after the last one.

HbA1c after a lapse

median HbA1c by months since first diabetic result

Never lapsed
Lapsed, then came back (results after return)

Show as table

Months since diagnosis	Stayed: median	n	Came back: median	n
0 to 3	8.2	1,372	–	0
3 to 6	7.8	509	–	0
6 to 9	7.5	393	–	1
9 to 12	7.3	271	8	63
12 to 15	7.2	243	8	89
15 to 18	7.2	168	8.3	103
18 to 21	7.1	127	8.2	145
21 to 24	7	82	8.2	109
24 to 27	7	67	8.1	83
27 to 30	7.1	45	8	76
30 to 33	6.8	29	7.8	54
33 to 36	7.1	19	7.9	44

People who came back after a lapse

n = 375 with an HbA1c before and on return

7.8%

median HbA1c before

8.5%

median HbA1c on return

15 mo

median time away

59%

came back 0.5+ points worse

The people the lab eventually sees again return worse off. Every month shaved off the gap is a month less of uncontrolled sugar.

First HbA1c band to latest

people with 2+ HbA1c at least 6 months apart, n = 1,522; row %

latest HbA1c (%)

< 5.75.7 to 6.46.5 to 7.9≥ 8.0

< 5.7

53%

31%

5.7 to 6.4

64%

27%

6.5 to 7.9

15%

68%

16%

≥ 8.0

58%

34%

rows: first HbA1c (%)

Most people stay in their band. The odd corner, a normal first HbA1c and a latest of 8 or more, is almost always a linking error rather than biology; see the plausibility check above.

Prediabetes: rechecked, and progressing

first result in the prediabetes range at least 15 months ago, n = 1,740

24%

rechecked within 15 months

419 people

38%

of those rechecked, now diabetic range

Prediabetes is where diabetes is cheapest to prevent, and most people with it never get the yearly recheck. They sit in the Routine tier of the call list, batched for a monthly SMS.

Month-end status of everyone diagnosed so far

people with a diabetic-range result, last 24 months

2,000

4,000

Oct 24Jan 25Apr 25Jul 25Oct 25Jan 26Apr 26Jul 26

On track or due soon
Overdue, still reachable (0 to 180 days)
Lost (over 180 days past due)

The lost pool grows every month (392 → 1,774), while the overdue-but-reachable band stays around 775. That middle band is where a call list pays off.

Show as table

Month end	On track	Overdue	Lost
31 Oct 2024	724	500	392
30 Nov 2024	708	533	445
31 Dec 2024	750	554	490
31 Jan 2025	762	568	553
28 Feb 2025	816	569	591
31 Mar 2025	807	591	658
30 Apr 2025	808	633	697
31 May 2025	806	657	771
30 Jun 2025	847	637	834
31 Jul 2025	847	677	890
31 Aug 2025	854	703	945
30 Sep 2025	887	690	1,023
31 Oct 2025	917	696	1,075
30 Nov 2025	932	698	1,150
31 Dec 2025	924	715	1,217
31 Jan 2026	930	721	1,283
28 Feb 2026	942	740	1,341
31 Mar 2026	932	766	1,408
30 Apr 2026	961	761	1,471
31 May 2026	987	771	1,537
30 Jun 2026	1,002	749	1,609
31 Jul 2026	1,001	769	1,663
31 Aug 2026	998	800	1,721
23 Sep 2026	1,087	797	1,774

How often people test, and whether it helps

The cohort you asked about: everyone with two or more sugar tests in the last 3.5 years. Testing rate is tests per year since a person's first test in the window, so both testing less and dropping out pull it down. The question is whether people who test more often stay in care and do better, or whether frequency just tracks who was already engaged.

2,863 people with 2+ tests3 tests each (median)1.5 tests / yr (median)

2,863

people with 2+ sugar tests in 3.5 years

60.3%

of people with diabetes tested 2+ times

2,206 people

+0.24

testing rate vs HbA1c improvement (Spearman ρ)

higher = frequent testers improve more · n = 1,805

-0.71

testing rate vs days overdue (ρ)

lower = frequent testers stay on time · n = 2,206

Do frequent testers stay in care?

% not yet lost to follow-up, by testing rate

Under 1 / yr32.7% n = 770
1 to 2 / yr57.2% n = 1,077
2 to 3 / yr82.4% n = 528
3+ / yr99% n = 488

Continuity climbs with testing rate, from 32.7% still in care in the least frequent group to 99% in the most frequent. Much of this is definitional, people who test often are by definition still testing, so read it alongside the retention curve rather than as a separate effect.

Control by testing frequency

median latest HbA1c, people with diabetes

7.6

Under 1 / yr

7.5

1 to 2 / yr

7.3

2 to 3 / yr

7.6

3+ / yr

Median HbA1c barely moves with testing rate, and the most frequent group is not the best controlled (at goal 34.1% in the least frequent group vs 21.1% in the most). Doctors repeat the test when sugar is high, so the sickest test most. The payoff is in the change over time and in staying in care, not in the level, see the improvement column below and the retention chart above.

Show as table

Testing rate	People (DM)	Median HbA1c	% at goal	Median change (pp)
Under 1 / yr	393	7.6	34.1%	+0.3
1 to 2 / yr	824	7.5	32.6%	+0.3
2 to 3 / yr	507	7.3	31.8%	+0.9
3+ / yr	482	7.6	21.1%	+1

Full yearly work-up by testing frequency

% of people with diabetes with 2+ HbA1c, eGFR and lipids in the last year

Under 1 / yr0% of 393
1 to 2 / yr4.1% of 824
2 to 3 / yr18.9% of 507
3+ / yr41.9% of 482

Frequent testers are far more likely to get the whole yearly panel, not just a repeat sugar. The kidney and lipid checks ride along with the habit of coming back.

What testing frequency tracks

Spearman ρ, people with diabetes

Latest HbA1cρ +0.01

no consistent link with lower HbA1c

HbA1c improvementρ +0.24

a moderate link: more testing goes with a bigger drop in HbA1c

Days past dueρ -0.71

a clear link: more testing goes with being less overdue

Association, not proof. Testing more does not by itself lower sugar; both are downstream of the same engagement, and a doctor orders repeat tests precisely when a patient is being actively managed. The useful read for the care team: the people testing under once a year are the ones to pull back in.

Who gets lost

Share of people lost within 12 months of their first diabetic-range result, among those diagnosed at least a year ago. Overall: 51.2%.

Contact on file

% lost within 12 months

Own mobile45% n = 2,084
Agent's number only78% n = 281
No usable number78% n = 203

First came through

% lost within 12 months

Walk-in, Barasat48% n = 1,335
Home collection48% n = 621
Partner centre61% n = 291
Screening camp62% n = 321

Age

% lost within 12 months

Under 4059% n = 261
40 to 5951% n = 1,421
60 and over50% n = 886

HbA1c at diagnosis

% lost within 12 months

6.5 to 7.944% n = 731
8.0 to 9.946% n = 883
10 and over46% n = 296

The cheapest fix on this page: record the patient’s own mobile at every registration, including home collections and camps. People without one are lost at 78% and 78%, against 45% for people with their own number. Until then, the call list routes them through the agent or referring doctor who can still reach them.

This week's call list

The rules below run every night on the linked timelines. The result is two queues: people who can still be kept, ranked by how sick and how late they are, and people already lost, for a slower re-engagement round. Each row carries the next step, who should act, and a ready-to-send reminder in English or Bengali. Open a name to see the timeline.

1,560 to keep in care2,961 to re-engagetop 150 + 50 loaded here

Keep-in-care queue by who acts

1,141Care coordinator

170Collection agent

127Referring doctor / outreach

122Duty doctor

WhyWho actsFind

150 match · top 150 of 1,560 in this queue loaded · 0 actioned · 0 booked

Critical88
P-04228 · F · 65yagent's number only
10.8%↑1.714 May 2026
Overdue 42 dayslast test 14 May 2026
Very high sugar, no repeatHbA1c overdue+2
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical85
P-04054 · M · 75y94xxx xx430
11.5%↑1.94 Mar 2026
Overdue 113 dayslast test 4 Mar 2026
HbA1c overdueHbA1c rising+2
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical85
P-06431 · F · 51y70xxx xx844
11.2%↑3.415 Apr 2026
Overdue 71 dayslast test 15 Apr 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical84
P-03780 · M · 35y96xxx xx439
12.7%↑1.221 Apr 2026
Overdue 65 dayslast test 21 Apr 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical83
P-06427 · M · 70yagent's number only
11.2%29 Jan 2026
Overdue 147 dayslast test 29 Jan 2026
HbA1c overdueKidney markers abnormal+1
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical83
P-01152 · F · 51y78xxx xx635
10.3%↑4.98 Feb 2026
Overdue 137 dayslast test 8 Feb 2026
HbA1c overdueHbA1c rising+2
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical81
P-02831 · F · 41y99xxx xx267
15.3%↑1.927 Feb 2026
Overdue 118 dayslast test 27 Feb 2026
HbA1c overdueHbA1c rising+2
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical81
P-06776 · M · 59y76xxx xx714
10.3%5 Apr 2026
Overdue 81 dayslast test 5 Apr 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical81
P-02422 · F · 63y68xxx xx349
11.0%↑1.515 May 2026
Overdue 41 dayslast test 15 May 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical80
P-06828 · F · 49y85xxx xx362
10.4%14 Apr 2026
Overdue 72 dayslast test 14 Apr 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical79
P-06487 · M · 70y98xxx xx113
10.3%↓0.419 Apr 2026
Overdue 67 dayslast test 19 Apr 2026
Very high sugar, no repeatHbA1c overdue+2
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical79
P-06701 · F · 39y79xxx xx653
12.3%26 Apr 2026
Overdue 60 dayslast test 26 Apr 2026
Very high sugar, no repeatHbA1c overdue+2
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor
Critical78
P-06406 · M · 78yagent's number only
10.1%26 Jan 2026
Overdue 150 dayslast test 26 Jan 2026
HbA1c overdueKidney check due+1
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical78
P-05967 · M · 74y95xxx xx925
10.1%↓0.612 Feb 2026
Overdue 133 dayslast test 12 Feb 2026
HbA1c overdueKidney check due+1
Doctor call first. Book HbA1c + eGFR + urine ACR + lipid profileDuty doctor
Critical78
P-06917 · M · 39y88xxx xx088
10.7%30 Apr 2026
Overdue 56 dayslast test 30 Apr 2026
Very high sugar, no repeatHbA1c overdue+3
Doctor callback within 48 h, repeat HbA1c + FBSDuty doctor

The rules behind the list

Intervals follow the ADA Standards of Care in Diabetes and RSSDI guidance. The weights are ours, written down so doctors can argue with them and change them in one place.

What is due, and when

Trigger	Next test	Basis
HbA1c 7% or higher (above goal)	every 90 days	ADA: quarterly until at goal
HbA1c under 7% (at goal)	every 180 days	ADA: at least twice a year when stable
One raised glucose, no HbA1c yet	within 30 days	Diagnosis needs a second abnormal result
HbA1c 10%+, FBS 250+ or PP 300+	doctor callback after 7 days with no repeat	Symptomatic range, treatment change likely
Diabetes: eGFR + urine ACR	yearly, 180 days if eGFR < 60 or ACR ≥ 30	ADA / KDIGO kidney screening
Diabetes: lipid profile	yearly (baseline within 90 days)	ADA cardiovascular risk review
Prediabetes (5.7 to 6.4%)	every 365 days	ADA: yearly recheck
No glycaemic test > 180 days past due	marked lost to follow-up	Moves to re-engagement, not routine recall

How the priority score is built

0 to 100

Factor	Points
Latest HbA1c ≥ 10 / 9 to 9.9 / 8 to 8.9 / 7 to 7.9 / < 7	40 / 32 / 24 / 16 / 8
Unconfirmed diabetic-range result	24
Very high sugar and no repeat within 7 days	+10, forces Critical
Days past due (peaks at 180, then decays)	up to 25
HbA1c up ≥ 1.0 / ≥ 0.5 since previous	10 / 5
Kidney markers abnormal / kidney check overdue	10 / 5
Lipid profile overdue	3
Diagnosed in the last 6 months	5
Age 65+	5

Critical 70+ (or very high sugar with no repeat), High 55 to 69, Medium 38 to 54, Routine below 38. Lateness peaks at 180 days and then decays: someone 3 months late is more recoverable than someone 3 years late.

Where it sits in the clinic's day

no new app for patients, no new data entry for staff

1AKTIV lab systembills + results, nightly mirror
2Link visitsphone, then fuzzy name
3Apply ruleswhat is due, how late, how risky
4Ranked queueskeep in care · re-engage
5Care team actscall, WhatsApp, agent, doctor

the next result for that person closes the item automatically, no manual tick-off

How the numbers were made

Data. Everything on this page is computed at build time from a synthetic cohort generated to look like the AKTIV to Neon research mirror: bills with a monthly PATIENT_KEY, names and phone numbers as a busy front desk types them, households sharing a mobile, and collection agents, camp desks and partner centres typing their own number. HbA1c drifts toward a treatment target while people keep coming and upward while they are away. No real patient appears here. The generator, the linker, the rules and the charts are separate modules, so the same code can be pointed at the mirror.

Linking. PATIENT_KEY is ignored for linking because it is reissued monthly. The mobile number is the identifier after cleaning; numbers with more than 15 different people are treated as shared (agent, camp or centre) numbers. Within a number, names are compared on a Bengali-aware phonetic key (Jaro-Winkler, first sound must agree), with sex and a birth-year tolerance. Visits with no usable number attach to a phone-linked person only when exactly one candidate matches, otherwise they link on full name, sex and age. Accuracy is scored against the generator’s hidden person id.

Definitions. Diabetic range: HbA1c ≥ 6.5%, fasting glucose ≥ 126 mg/dL, or post-meal / random ≥ 200 mg/dL. A single raised glucose with no HbA1c is “unconfirmed” and gets a confirmatory test. Prediabetes: HbA1c 5.7 to 6.4%, fasting 100 to 125, post-meal 140 to 199. Next HbA1c is due 90 days after a result ≥ 7% and 180 days after a result under 7%. Lost to follow-up: more than 180 days past due with no sugar test.

What this cannot see. A lab only sees tests done at the lab. Someone “lost” here may be testing elsewhere, have moved, or have died, which is why the re-engagement queue is kept separate and its first question is “where are you getting tested now?”. Medicines, symptoms and clinic visits are not in the mirror. Scoring weights are a starting point for the clinical team, not a validated model.