How the Vårdbetyg score is calculated

Everything above applies to health centres. The site also compares specialist outpatient clinics (specialistmottagningar) and emergency departments (akutmottagningar), and there the model differs for a concrete reason: the National Board of Health and Welfare’s waiting-time statistics cover primary care and cannot be applied to the other verticals. Those scores therefore rest on patient experience only. Both verticals are published in Swedish only.

• Specialist outpatient clinics – the patient survey’s measurement of specialised outpatient hospital care, whose leading dimension is Helhetsintryck (overall impression) rather than Vård och behandling. The same Bayesian shrinkage as above, but towards that measurement’s own national average. It runs every second year; the latest survey year is 2025, and regions that sat out that round are shown with their own survey year. The waiting time to a first visit is shown as context beside the score, never inside it.
• Emergency departments – the patient survey’s emergency measurement, run every second year on even years; the latest survey year is 2024, and a region that sat out that round is shown with its own survey year. The National Board’s emergency waiting times are measured per hospital, not per department, and the registers count different populations. They therefore appear as context on the hospital page and are never weighed into an individual department’s score.

In both verticals we compare only within a region. Regions report at different organisational levels, one per clinic and another per department group, so a national leaderboard would compare unlike things. For the same reason there are no awards and no seals there: the awards above apply to health centres only.

The honesty rules are identical everywhere: shrinkage towards the national average, a response floor, and a plain “too few responses” rather than an invented figure. In the specialist and emergency measurements the patient survey withholds units under 30 responses altogether, so there the floor is set by the survey itself.

Beyond the score itself we highlight the health centres that truly stand out. Three awards can be earned – all derived entirely from the same open data as the score and impossible to buy. Each award carries the award year (the survey year plus one) and always cites the survey as its source. It is set when the survey is released and holds for the whole award year: monthly waiting-time updates never change who holds it, reassessment happens at the next survey period, and a unit that closes loses its award immediately:

• Top-rated in the municipality – the health centre has the highest Vårdbetyg score in its municipality. The award requires the municipality to have at least 3 rated health centres, so that first place actually means something. If two health centres have exactly the same score, the award goes to the one that comes first alphabetically.
• Top 10% in Sweden – the health centre’s Vårdbetyg score is among the highest 10 percent in the country. The boundary is inclusive: if several health centres sit exactly on the cutoff value, all of them qualify, so the share can slightly exceed 10 percent. Better that than an arbitrary tie-break at the boundary.
• Most improved – the health centre’s patient responses have improved over time, measured on the five areas covered by both the earlier and the current survey. Two conditions apply: the improvement is larger than the margin of error, and larger than Sweden’s change over the same years. The verdict is only made within one and the same questionnaire, never across the 2024/2025 change, and requires at least 3 survey years, at least 30 responses in the last of them and a current score. As long as the new questionnaire has only one survey year, the award reflects improvement in the earlier survey, up to and including 2024, and the span is always stated next to the mark. The award says something about direction, not level, which is why the score is always shown beside it. There is no counterpart in the other direction.

Setting an award requires a more solid basis than the score itself: at least 30 responses, against 10 to get a score at all. Highlighting a health centre as one of the best demands more certainty than simply showing a number. The Bayesian shrinkage above already pulls units with a thin basis towards the national average, so that no one should be able to top a list by chance – the response floor is an extra, clear safeguard.

The awards are verifiable. Every awarded health centre’s page shows the score and the number of responses behind it, and its place in the municipality when the municipality has enough rated health centres. No national placement is shown, because the top award rests on a threshold for the highest 10 percent rather than on a ranking. We never charge for placements, and no health centre can influence its award in any way other than through better results.

On some health centres’ pages we show the share of listed patients who have a named regular doctor (fast läkarkontakt). It is one of the strongest quality signals in primary care, but it is not part of the Vårdbetyg score – it stands as descriptive context beside the score.

The measure comes from released public records (data requests) from three regions, and is site-exclusive: it never appears in our open data export. The three regions count in similar but not identical ways, so the denominator differs slightly:

• Skåne – the region’s own quarterly share (patients listed on a named doctor divided by total listed), the unit’s most recent stable quarter through Q4 2025.
• Stockholm – the GP listing, patients listed on a named doctor divided by the total. The basis is three extracts (October 2025, January 2026 and April 2026), and each unit is shown with its stable median extract.
• Västra Götaland – monthly extracts January–July 2026: the sum of patients listed on the unit’s named doctors divided by total listed. VGR structurally lists nearly everyone on a named doctor, and the doctor register and the total list have different extract vintages, so the share is capped at 100% (a patient cannot have more than one regular doctor; the excess is register noise).

Compare confidently within a region, but carefully between regions – the definitions are not normalised into a single national measure, and coverage is three regions, not the whole country. An artefact guard picks the unit’s most recent stable period and excludes obvious register gaps (for example near-zero patients listed on a doctor); those units get no figure rather than a misleading one.

A low share can also reflect how the clinic registers listings, and a unit with many elderly or chronically ill patients can have a structurally higher share. Letting the measure affect the score itself would be a methodological and product decision that requires more even coverage; today it is pure context.

On Skåne health centres’ pages we also show the region’s continuity index (quality indicator 3.2 in the Hälsoval Skåne follow-up): the share of patients who saw the same doctor more than half of their visits, among patients with more than three physical doctor visits in a rolling six-month window. Where named GP contact measures who you are listed with, the continuity index measures who patients actually saw – the continuity research links to fewer emergency visits and lower mortality.

The measure comes from a released public record (a data request against the region’s follow-up system) and is site-exclusive: it never appears in our open data export, and it is not part of the Vårdbetyg score.

The per-unit sample is small: only patients with more than three visits in six months count, typically a few dozen per health centre. We therefore only show the figure when at least 30 patients are included, and always print how many patients it is based on. Units below the floor get no figure rather than a misleading one. Coverage is currently one region (Skåne), so compare within the region.

On every health centre’s page we show how patients’ ratings have developed 2015–2024, per dimension, as small curves with the latest value written out. The series comes from the earlier NPE measurement (primary care 2015–2024), which uses different dimensions and a different design than the 2025 survey. That is why we never merge the history with today’s ratings: it stands beside the score as context, not inside it. For the dimensions that also exist in the 2025 survey, 2025 is shown as a separate point after a marked break in the curve, because the levels are not comparable across the change of survey methodology. Units are matched across years via their HSA id, the only key that is stable across the measurements.

The history follows the same honesty rules as the rest of the site:

• Years without results are shown as breaks in the curve – we never interpolate values that do not exist.
• Uncertainty is drawn: the band around the curve is the survey’s margin of error (confidence interval). Where the interval is missing the band breaks, rather than pretending the uncertainty is zero.
• A unit with only a single result gets a text note instead of a curve – one point is not a trend.

The verdicts next to the curves, such as “Better 2015–2024”, are always tested against what chance alone could explain (the margin of error) and always calculated within one and the same survey system. The 2025 survey has new questions and response scales and is a new starting point, a new time series, so 2025 is never compared with earlier years and never enters a verdict. Only when the new series has more vintages can a direction be tested within it.

The model’s key choices are not matters of taste. Weighing patients’ answers heavily, treating continuity as worth examining and shrinking small samples are all supported by published research. Here are the key sources, so you can review them yourself.

Patients’ answers measure real quality. A systematic review of 55 studies found consistent associations between patients’ experience of care and both patient safety and clinical effectiveness (Doyle, Lennox & Bell 2013, BMJ Open). Survey answers are not a popularity contest, but a measurable side of care quality.

Continuity has the strongest research support in primary care. That is why the site shows a named regular doctor as its own context where regional data exists, and why continuity is one of the survey dimensions in the score.

• In Norway, whose regular-GP system resembles what Sweden is trying to build, 4.5 million residents were followed: people who had kept the same doctor for more than 15 years had roughly 25 per cent lower mortality and 28 per cent fewer acute admissions than those with a brand-new doctor relationship (Sandvik et al. 2022, British Journal of General Practice).
• In Denmark, where patients register with a clinic just as in Sweden, longer time with the same clinic meant lower mortality and fewer unplanned hospital contacts in a study of the entire adult population (The Lancet Primary Care 2025).
• A systematic review found that 18 of 22 studies show significantly lower mortality with higher doctor continuity (Pereira Gray et al. 2018, BMJ Open), and Sweden’s HTA agency SBU reached the same conclusion in its assessment (SBU 2021, Continuity of care).
• Sweden also ranks last in an international comparison: 32 per cent of Swedes have a named regular doctor, against an average of 81 per cent across the ten countries compared (Swedish Agency for Health and Care Services Analysis, IHP 2023).

Why we shrink small samples. That naive league tables mislead, because rankings are extremely sensitive to chance when observations are few, was shown by Goldstein & Spiegelhalter as early as 1996 (League Tables and Their Limitations, Journal of the Royal Statistical Society). Its successor is standard in British healthcare statistics: uncertainty should govern how much weight a unit’s deviation carries (Spiegelhalter 2005, Statistics in Medicine). The shrinkage and confidence intervals above are that principle, applied to Swedish health centres.

Primary care at large. That a well-functioning primary care system produces better and more evenly distributed health is established in the research review the field rests on (Starfield, Shi & Macinko 2005, The Milbank Quarterly). Choosing a health centre, in other words, is a decision worth taking seriously.

A note on honesty: effect sizes from other countries should be read as direction, not as Swedish forecasts. That is why we lead with Nordic studies, whose systems resemble ours, and use the wider international research as support for the pattern.

The Vårdbetyg score is not adjusted for who the registered patients are. Age, language and socioeconomic conditions differ between health centres’ catchment areas, and that affects how patients answer surveys. You should know this when comparing.

The research is clear on this point. A Swedish government agency analysis of the National Patient Survey itself found that patient-reported quality is lower in big cities and at health centres whose registered patients face tougher socioeconomic conditions (Swedish Agency for Health and Care Services Analysis 2012:1). An English methods study of over two million survey responses quantifies the effect: without adjustment, practices serving more deprived populations are disadvantaged, typically by 5 to 10 points on a 100-point scale (Paddison et al. 2012, BMJ Quality & Safety).

In practice this means: a health centre in a disadvantaged area with a slightly lower score may be doing at least as good a job as a higher-scoring unit in an affluent area. The regions’ own funding of health centres compensates for this through socioeconomic indices, but survey results do not, and therefore neither does our score.

The same analysis showed that the difference between privately and publicly run health centres largely disappeared after adjusting for socioeconomics and care needs. That is why we present ownership as neutral fact on each unit’s page and never use it as a quality argument.

Why do we not adjust the score? Such an adjustment is a major methodological choice that must be made openly and be open to scrutiny, and no Swedish comparison service makes it today. Until we possibly do, we choose to state the limitation plainly instead, right here.