Why English? | Reporting PRISMA-Harms Reviews Through Adverse Events, Zero Counts, Ascertainment and Causality

PRISMA-Harms systematic review reporting matters because safety information is often rarer, less consistently defined and collected differently from benefit outcomes. People searching for adverse events, side effects, zero-event studies or harms meta-analysis need to know what was actively measured, what was passively reported and what may have been missed.

The official PRISMA-Harms page, checked on 5 November 2026, says the checklist applies to any systematic review addressing harms, whether harms are analysed alone or with benefits. The open 2016 statement identifies four essential extensions: mention harms in the title, specify handling of zero events, define and describe ascertainment and time period for each harm, and describe any assessment of possible causality. A later BMJ correction switched two definitions in the original table, so this article relies on the corrected interpretation rather than repeating the error.

English is part of safety. Adverse event does not automatically mean an intervention caused the event. “No serious events” can mean zero events were observed, that events were not actively sought, or that reporting was incomplete. Clear prose keeps severity, frequency, timing, causality and denominator separate.

Use this guide with eduKate’s ROB-ME missing-evidence guide and GRADE certainty guide. Complete reporting helps readers see the evidence; it does not remove sparse data, missing results or uncertainty.

---

State the Harms Question in the Title and Objective

Name the harm or harms focus so readers can find and interpret the review correctly.

Harms reporting task

The official PRISMA-Harms extension applies whenever a systematic review addresses harms, whether harms are primary or considered with benefits. For state the harms question in the title and objective, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

One trial reports no serious events after active monthly questioning, while another says nothing about how events were sought. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

PRISMA-Harms is a reporting extension, not a substitute for clinical judgement, pharmacovigilance or regulatory evidence. Under state the harms question in the title and objective, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Define Adverse Event, Adverse Effect and Serious Event

Use operational definitions and avoid implying causality from timing alone.

Read benefit and harm on compatible terms

Four essential extensions concern identifying harms in the title, handling zero events, defining and ascertaining each harm over time, and assessing possible causality. In define adverse event, adverse effect and serious event, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

A review excludes studies with zero events in both groups, reducing the visible evidence about rare outcomes. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

A reported event is not automatically caused by the intervention. Finish define adverse event, adverse effect and serious event with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Prespecify Populations, Exposures, Doses and Time Windows

Harms can vary with who is exposed, how much, for how long and alongside what else.

Follow every event to its denominator

Harms terminology should distinguish adverse events from events judged causally related to an intervention. For prespecify populations, exposures, doses and time windows, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

Headache counts include repeated episodes in some participants, but the analysis treats every event as if it came from a different person. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

Zero observed events do not prove zero risk, especially with small samples or short follow-up. Under prespecify populations, exposures, doses and time windows, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Register Harms Methods Before Results

Plan search terms, eligible sources, event groupings, ascertainment and sparse-data methods.

Harms reporting task

Searches may need sources and terms that capture adverse effects, withdrawals, complications and long-term follow-up. For register harms methods before results, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

A withdrawal is attributed to treatment by an investigator, while another study records only the timing and makes no causality judgement. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

Harms may vary with dose, duration, population and co-interventions. Under register harms methods before results, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Search Specifically for Harms Evidence

Use synonyms, adverse-event terms, regulatory or registry sources and longer follow-up where relevant.

Read benefit and harm on compatible terms

Passive reporting, active questioning, checklists and record review can yield different event frequencies. In search specifically for harms evidence, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

Short trials capture immediate symptoms but cannot answer a question about delayed harm. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

Combining unlike event definitions can create a misleading total. Finish search specifically for harms evidence with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Include Designs Suited to Rare or Delayed Events

Explain when non-randomised or surveillance evidence complements trials.

Follow every event to its denominator

Sparse and zero-event data require transparent analytic choices because exclusions can distort an estimate. For include designs suited to rare or delayed events, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

A forest plot shows a wide interval compatible with meaningful benefit and meaningful harm despite a neutral point estimate. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

Individual decisions require current, person-specific advice where safety is at stake. Under include designs suited to rare or delayed events, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Track Reports, Studies and Follow-Up Periods

Multiple publications from one study can otherwise double-count participants or events.

Harms reporting task

Study-level tables should preserve the event definition, collection method, follow-up period and denominator. For track reports, studies and follow-up periods, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

Families compare a programme’s reported incidents with how actively the programme looked for and recorded them. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

Benefit and harm should be presented on compatible time horizons and denominators where possible. Under track reports, studies and follow-up periods, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Define Every Harm Before Extraction

State severity, diagnostic criteria, coding, timing and whether repeated events are possible.

Read benefit and harm on compatible terms

Conclusions should balance benefits and harms without treating non-significance as proof of safety. In define every harm before extraction, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

Different definitions of a serious event are harmonised transparently rather than silently merged. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

The correction to the 2016 paper’s definitions should be noted when consulting the original table. Finish define every harm before extraction with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Report How Each Harm Was Ascertained

Distinguish active questioning, checklists, spontaneous reports, clinician judgement and record review.

Follow every event to its denominator

The official PRISMA-Harms extension applies whenever a systematic review addresses harms, whether harms are primary or considered with benefits. For report how each harm was ascertained, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

One trial reports no serious events after active monthly questioning, while another says nothing about how events were sought. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

PRISMA-Harms is a reporting extension, not a substitute for clinical judgement, pharmacovigilance or regulatory evidence. Under report how each harm was ascertained, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Keep Participants, Events and Withdrawals Separate

A person with several events is not the same denominator as several people with one event.

Harms reporting task

Four essential extensions concern identifying harms in the title, handling zero events, defining and ascertaining each harm over time, and assessing possible causality. For keep participants, events and withdrawals separate, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

A review excludes studies with zero events in both groups, reducing the visible evidence about rare outcomes. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

A reported event is not automatically caused by the intervention. Under keep participants, events and withdrawals separate, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Record Observation Time and Exposure

Rates per person-time and cumulative risks answer different questions.

Read benefit and harm on compatible terms

Harms terminology should distinguish adverse events from events judged causally related to an intervention. In record observation time and exposure, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

Headache counts include repeated episodes in some participants, but the analysis treats every event as if it came from a different person. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

Zero observed events do not prove zero risk, especially with small samples or short follow-up. Finish record observation time and exposure with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Preserve Zero-Event Studies

Explain whether and how studies with zero events in one or both groups enter the analysis.

Follow every event to its denominator

Searches may need sources and terms that capture adverse effects, withdrawals, complications and long-term follow-up. For preserve zero-event studies, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

A withdrawal is attributed to treatment by an investigator, while another study records only the timing and makes no causality judgement. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

Harms may vary with dose, duration, population and co-interventions. Under preserve zero-event studies, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Choose Sparse-Data Methods Transparently

Name continuity corrections, exact methods or models and test important alternatives.

Harms reporting task

Passive reporting, active questioning, checklists and record review can yield different event frequencies. For choose sparse-data methods transparently, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

Short trials capture immediate symptoms but cannot answer a question about delayed harm. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

Combining unlike event definitions can create a misleading total. Under choose sparse-data methods transparently, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Avoid Treating Non-Significance as Safety

Wide intervals can include important harm even when the point estimate appears neutral.

Read benefit and harm on compatible terms

Sparse and zero-event data require transparent analytic choices because exclusions can distort an estimate. In avoid treating non-significance as safety, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

A forest plot shows a wide interval compatible with meaningful benefit and meaningful harm despite a neutral point estimate. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

Individual decisions require current, person-specific advice where safety is at stake. Finish avoid treating non-significance as safety with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Assess Heterogeneity in Definitions and Collection

Different surveillance intensity can create apparent differences in event frequency.

Follow every event to its denominator

Study-level tables should preserve the event definition, collection method, follow-up period and denominator. For assess heterogeneity in definitions and collection, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

Families compare a programme’s reported incidents with how actively the programme looked for and recorded them. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

Benefit and harm should be presented on compatible time horizons and denominators where possible. Under assess heterogeneity in definitions and collection, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Report Individual-Study Harms With Denominators

Tables should allow readers to reconstruct events, people at risk and follow-up.

Harms reporting task

Conclusions should balance benefits and harms without treating non-significance as proof of safety. For report individual-study harms with denominators, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

Different definitions of a serious event are harmonised transparently rather than silently merged. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

The correction to the 2016 paper’s definitions should be noted when consulting the original table. Under report individual-study harms with denominators, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Describe Any Causality Assessment

State who judged relatedness, which criteria were used and how disagreements were handled.

Read benefit and harm on compatible terms

The official PRISMA-Harms extension applies whenever a systematic review addresses harms, whether harms are primary or considered with benefits. In describe any causality assessment, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

One trial reports no serious events after active monthly questioning, while another says nothing about how events were sought. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

PRISMA-Harms is a reporting extension, not a substitute for clinical judgement, pharmacovigilance or regulatory evidence. Finish describe any causality assessment with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Consider Selective Non-Reporting

Compare protocols, registries and publications for missing harm outcomes or analyses.

Follow every event to its denominator

Four essential extensions concern identifying harms in the title, handling zero events, defining and ascertaining each harm over time, and assessing possible causality. For consider selective non-reporting, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

A review excludes studies with zero events in both groups, reducing the visible evidence about rare outcomes. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

A reported event is not automatically caused by the intervention. Under consider selective non-reporting, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Balance Benefits and Harms on Compatible Terms

Use comparable populations and time horizons and show absolute effects where possible.

Harms reporting task

Harms terminology should distinguish adverse events from events judged causally related to an intervention. For balance benefits and harms on compatible terms, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

Headache counts include repeated episodes in some participants, but the analysis treats every event as if it came from a different person. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

Zero observed events do not prove zero risk, especially with small samples or short follow-up. Under balance benefits and harms on compatible terms, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Did You Know? Looking Harder Can Find More Events

Active surveillance may produce higher counts because detection is better, not because care is worse.

Read benefit and harm on compatible terms

Searches may need sources and terms that capture adverse effects, withdrawals, complications and long-term follow-up. In did you know? looking harder can find more events, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

A withdrawal is attributed to treatment by an investigator, while another study records only the timing and makes no causality judgement. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

Harms may vary with dose, duration, population and co-interventions. Finish did you know? looking harder can find more events with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Translate Harms Evidence for Students and Parents

Ask what was measured, how long, how actively and with how much uncertainty.

Follow every event to its denominator

Passive reporting, active questioning, checklists and record review can yield different event frequencies. For translate harms evidence for students and parents, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

Short trials capture immediate symptoms but cannot answer a question about delayed harm. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

Combining unlike event definitions can create a misleading total. Under translate harms evidence for students and parents, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Know When Professional Advice Is Needed

Population evidence cannot replace current person-specific safety guidance.

Harms reporting task

Sparse and zero-event data require transparent analytic choices because exclusions can distort an estimate. For know when professional advice is needed, define the event, ascertainment method, observation window and denominator before comparing groups. English matters because adverse event, adverse effect, side effect, complication, withdrawal and serious event do not automatically mean the same thing or imply the same causal relationship.

Worked harms example

A forest plot shows a wide interval compatible with meaningful benefit and meaningful harm despite a neutral point estimate. Separate “no events occurred” from “events were not sought” and “events were not reported”. Then show how zero counts entered the analysis. A table should preserve participants at risk, study duration, collection method and uncertainty; a confident sentence based on sparse passive reporting can hide more than it reveals.

Safety boundary and next action

Individual decisions require current, person-specific advice where safety is at stake. Under know when professional advice is needed, report the supported signal without diagnosing an individual or promising safety. A review can identify patterns, limitations and research needs, but personal decisions may require current professional advice and patient-specific information. The practical next step may be checking protocols, regulatory sources, longer follow-up or active surveillance methods.

---

Write a Proportionate Conclusion

Separate a signal, a plausible association, insufficient evidence and supported causality.

Read benefit and harm on compatible terms

Study-level tables should preserve the event definition, collection method, follow-up period and denominator. In write a proportionate conclusion, do not let a detailed benefit analysis sit beside one vague sentence about safety. State the populations, time horizons and data sources for both. Harms may be rarer, delayed, inconsistently defined or collected outside the primary outcome schedule, so the review needs methods designed for those features.

Ascertainment and causality

Families compare a programme’s reported incidents with how actively the programme looked for and recorded them. Ask whether participants were questioned actively, invited to report spontaneously, assessed with a checklist or identified from records. These approaches can produce different event counts. If authors judge an event related to an intervention, report who made the judgement and how; temporal sequence alone is not proof of causality.

Interpret uncertainty honestly

Benefit and harm should be presented on compatible time horizons and denominators where possible. Finish write a proportionate conclusion with absolute numbers and intervals where possible. Absence of a statistically clear difference is not proof of no harm, especially with few events. Explain whether the evidence is reassuring, concerning or simply insufficient, and keep severity, reversibility and frequency separate.

---

Finish With a PRISMA-Harms Audit

Verify title, definitions, ascertainment, time, denominator, zero events, causality, missing evidence and balanced conclusions.

Follow every event to its denominator

Conclusions should balance benefits and harms without treating non-significance as proof of safety. For finish with a prisma-harms audit, trace the path from eligibility and search through event definition, extraction, synthesis and conclusion. A review should not mix event counts, numbers of people with at least one event and withdrawals without explaining the unit. Repeat events in one person also require a suitable analytic approach.

Example and zero-event decision

Different definitions of a serious event are harmonised transparently rather than silently merged. State whether studies with zero events in one or both groups were retained, excluded or handled with a particular model. Different choices can change a sparse-data result. The report should explain sensitivity analyses and avoid continuity corrections or transformations that appear only after the result is known.

Practical reading move

The correction to the 2016 paper’s definitions should be noted when consulting the original table. Under finish with a prisma-harms audit, ask what was measured, for how long, by whom and with what chance of missing an event. Parents and students can use the same language habit when reading product, school or activity safety information: distinguish documented events from surveillance quality and avoid turning a population estimate into a personal guarantee.

---

Frequently Asked Questions

What is PRISMA-Harms?

A reporting extension for systematic reviews that address harms, whether harms are primary or considered with benefits.

What are the four essential extensions?

Identify harms in the title; explain zero-event handling; define harms, ascertainment and time periods; and describe causality assessment.

Is an adverse event always caused by the intervention?

No. An adverse event occurs during or after exposure; causal attribution requires additional judgement and evidence.

What does zero events mean?

It may mean none were observed under the study’s methods and time window. It does not prove zero risk.

Why does ascertainment matter?

Active questioning, spontaneous reports, checklists and records can detect different numbers and types of events.

Can studies with no events be excluded?

Exclusion can distort sparse evidence. The review should state and justify its analytic handling.

Does no statistically significant difference prove safety?

No. Precision, event counts, interval width and surveillance quality must be considered.

Does the checklist replace medical or regulatory advice?

No. It improves reporting; personal safety decisions may require current professional and regulatory information.

---

Useful Next Reading

• Read the official PRISMA-Harms page and checklist.
• Read the open PRISMA-Harms statement.
• Examine missing evidence with ROB-ME.
• Connect harms estimates to GRADE certainty.

Closing thought: PRISMA-Harms shows why English is a safety tool: definitions, denominators and careful causal words keep a sparse event record from becoming false reassurance or unnecessary alarm.